ai-agent-book 精选快照（<2MB 代码与文档，来自 github.com/bojieli/ai-agent-book）
Build latest book artifacts / build (push) Canceled after 0s
dependency resolution / resolve (3.11) (push) Canceled after 0s
dependency resolution / resolve (3.13) (push) Canceled after 0s
deploy-pages / build (push) Canceled after 0s
deploy-pages / deploy (push) Canceled after 0s
i18n consistency check / check (push) Canceled after 0s
provider adoption tests / test (chapter2/context-compression) (push) Canceled after 0s
provider adoption tests / test (chapter2/prompt-injection) (push) Canceled after 0s
provider adoption tests / test (chapter2/system-hint) (push) Canceled after 0s
provider adoption tests / test (chapter3/log-sanitization) (push) Canceled after 0s
web-search-agent tests / test (push) Canceled after 0s
web-search-agent tests / agentbook (push) Canceled after 0s
Build latest book artifacts / build (push) Canceled after 0s
dependency resolution / resolve (3.11) (push) Canceled after 0s
dependency resolution / resolve (3.13) (push) Canceled after 0s
deploy-pages / build (push) Canceled after 0s
deploy-pages / deploy (push) Canceled after 0s
i18n consistency check / check (push) Canceled after 0s
provider adoption tests / test (chapter2/context-compression) (push) Canceled after 0s
provider adoption tests / test (chapter2/prompt-injection) (push) Canceled after 0s
provider adoption tests / test (chapter2/system-hint) (push) Canceled after 0s
provider adoption tests / test (chapter3/log-sanitization) (push) Canceled after 0s
web-search-agent tests / test (push) Canceled after 0s
web-search-agent tests / agentbook (push) Canceled after 0s
This commit is contained in:
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,355 @@
|
||||
"""
|
||||
Prompt 蒸馏「蒸馏前 vs 蒸馏后」量化对比脚本。
|
||||
|
||||
本脚本回答实验 8-8 的核心问题：把「长提示 + 思考型教师」蒸馏成「无提示 + 直接
|
||||
回答的学生」之后，到底省了多少、质量掉了多少？它在 **不加载任何大模型、不联网** 的
|
||||
前提下，用真实数据算出一张 before/after 对比表：
|
||||
|
||||
1. 输入成本（token）：教师每次调用都要带上完整的语言分类提示（约上千 token），
|
||||
学生只需要原始待分类文本。二者的 token 差就是每次调用省下的输入开销。
|
||||
2. 任务质量：直接读取 evaluate.py 产出的 evaluation_results.json，得到学生在
|
||||
相同输入上「与教师标注的一致率」（即蒸馏保真度）。
|
||||
3. 逐条案例：抽取若干条真实样本，并排展示 教师 token / 学生 token / 教师标签 /
|
||||
学生预测 / 是否一致，让「多个案例上的 before/after」一目了然。
|
||||
|
||||
设计原则：所有数字都来自真实数据与真实分词器，不臆造。延迟（秒级响应时间）需要
|
||||
在 GPU 上实测，本脚本不做估算，只报告可离线复现的 token 成本与质量。
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
import sys
|
||||
from pathlib import Path
|
||||
from typing import Callable, Dict, List, Optional, Tuple
|
||||
|
||||
|
||||
VALID_LABELS = ["ar", "de", "el", "en", "es", "fr", "hi", "ru", "tr", "ur", "vi", "zh", "ot"]
|
||||
|
||||
|
||||
def load_prompt_template(source_file: str) -> str:
|
||||
"""从 create_data.py 中提取教师使用的语言分类提示模板（避免 import vllm）。"""
|
||||
src = Path(source_file).read_text(encoding="utf-8")
|
||||
match = re.search(
|
||||
r'LANGUAGE_CLASSIFICATION_PROMPT\s*=\s*"""(.*?)"""',
|
||||
src,
|
||||
re.DOTALL,
|
||||
)
|
||||
if not match:
|
||||
raise ValueError(
|
||||
f"无法在 {source_file} 中找到 LANGUAGE_CLASSIFICATION_PROMPT 模板，"
|
||||
f"请用 --prompt_source 指定包含该常量的文件。"
|
||||
)
|
||||
return match.group(1)
|
||||
|
||||
|
||||
def build_token_counter(tokenizer_name: Optional[str]) -> Tuple[Callable[[str], int], str]:
|
||||
"""
|
||||
构造一个 token 计数函数，按优先级回退，保证离线可用。
|
||||
|
||||
返回 (counter, method_description)：
|
||||
1) 若指定 --tokenizer，用 HuggingFace 分词器精确计数（GPU 机器上可得到 Qwen 的真实 token 数）。
|
||||
2) 否则用 tiktoken 的 o200k_base（GPT-4o/o1 分词器）作近似，可离线复现。
|
||||
3) 再退化为「字符数 / 4」的粗略启发式，并明确标注为估算。
|
||||
每种方法都会在输出里注明，绝不把近似值当成精确值。
|
||||
"""
|
||||
if tokenizer_name:
|
||||
try:
|
||||
from transformers import AutoTokenizer
|
||||
|
||||
tok = AutoTokenizer.from_pretrained(tokenizer_name, trust_remote_code=True)
|
||||
return (lambda s: len(tok.encode(s))), f"HuggingFace 分词器（精确）: {tokenizer_name}"
|
||||
except Exception as exc: # noqa: BLE001
|
||||
print(f"[warn] 无法加载分词器 {tokenizer_name}（{exc}），回退到 tiktoken。", file=sys.stderr)
|
||||
|
||||
try:
|
||||
import tiktoken
|
||||
|
||||
enc = tiktoken.get_encoding("o200k_base")
|
||||
return (lambda s: len(enc.encode(s))), "tiktoken o200k_base（近似，可离线复现）"
|
||||
except Exception as exc: # noqa: BLE001
|
||||
print(f"[warn] tiktoken 不可用（{exc}），回退到字符启发式。", file=sys.stderr)
|
||||
|
||||
return (lambda s: max(1, len(s) // 4)), "字符数/4（粗略估算）"
|
||||
|
||||
|
||||
def load_texts(test_file: str) -> List[str]:
|
||||
with open(test_file, "r", encoding="utf-8") as f:
|
||||
return [line.strip() for line in f if line.strip()]
|
||||
|
||||
|
||||
def load_teacher_labels(train_data_file: str) -> Dict[str, str]:
|
||||
"""从蒸馏训练数据（教师标注）中读取 文本 -> 教师标签 的映射。"""
|
||||
mapping: Dict[str, str] = {}
|
||||
if not Path(train_data_file).exists():
|
||||
return mapping
|
||||
with open(train_data_file, "r", encoding="utf-8") as f:
|
||||
for line in f:
|
||||
if not line.strip():
|
||||
continue
|
||||
data = json.loads(line)
|
||||
msgs = data.get("messages", [])
|
||||
if len(msgs) >= 2:
|
||||
mapping[msgs[0].get("content", "")] = msgs[1].get("content", "")
|
||||
return mapping
|
||||
|
||||
|
||||
def load_eval_results(eval_file: str) -> Optional[Dict]:
|
||||
if not Path(eval_file).exists():
|
||||
return None
|
||||
with open(eval_file, "r", encoding="utf-8") as f:
|
||||
return json.load(f)
|
||||
|
||||
|
||||
def truncate(text: str, width: int = 42) -> str:
|
||||
text = text.replace("\n", " ")
|
||||
return text if len(text) <= width else text[: width - 1] + "…"
|
||||
|
||||
|
||||
def compare(
|
||||
prompt_template: str,
|
||||
texts: List[str],
|
||||
teacher_labels: Dict[str, str],
|
||||
eval_results: Optional[Dict],
|
||||
count_tokens: Callable[[str], int],
|
||||
token_method: str,
|
||||
num_examples: int,
|
||||
) -> Dict:
|
||||
n = len(texts)
|
||||
|
||||
# 固定提示开销（模板本身，不含待分类文本）
|
||||
fixed_overhead = count_tokens(prompt_template.format(text=""))
|
||||
|
||||
teacher_input_total = 0
|
||||
student_input_total = 0
|
||||
per_text_tokens: List[Tuple[int, int]] = [] # (teacher_tokens, student_tokens)
|
||||
for text in texts:
|
||||
teacher_prompt = prompt_template.format(text=text)
|
||||
t_tok = count_tokens(teacher_prompt)
|
||||
s_tok = count_tokens(text)
|
||||
teacher_input_total += t_tok
|
||||
student_input_total += s_tok
|
||||
per_text_tokens.append((t_tok, s_tok))
|
||||
|
||||
if n == 0:
|
||||
teacher_avg = student_avg = reduction_pct = 0.0
|
||||
ratio = float("inf")
|
||||
else:
|
||||
teacher_avg = teacher_input_total / n
|
||||
student_avg = student_input_total / n
|
||||
reduction_pct = (
|
||||
100.0 * (1 - student_input_total / teacher_input_total)
|
||||
if teacher_input_total
|
||||
else 0.0
|
||||
)
|
||||
ratio = (
|
||||
teacher_input_total / student_input_total
|
||||
if student_input_total
|
||||
else float("inf")
|
||||
)
|
||||
|
||||
# 学生预测（与 test_file 逐行对齐）
|
||||
student_preds: Optional[List[Optional[str]]] = None
|
||||
accuracy = None
|
||||
correct = evaluated = None
|
||||
if eval_results:
|
||||
student_preds = eval_results.get("predictions")
|
||||
summary = eval_results.get("summary", {})
|
||||
accuracy = summary.get("accuracy")
|
||||
correct = summary.get("correct")
|
||||
evaluated = summary.get("evaluated")
|
||||
|
||||
# 逐条案例：优先覆盖不同语言，并尽量各带上一致/不一致的例子
|
||||
examples: List[Dict] = []
|
||||
seen_labels = set()
|
||||
for idx, text in enumerate(texts):
|
||||
teacher_label = teacher_labels.get(text, "?")
|
||||
student_pred = (
|
||||
student_preds[idx] if student_preds and idx < len(student_preds) else None
|
||||
)
|
||||
key = teacher_label
|
||||
if key in seen_labels and len(examples) >= num_examples:
|
||||
continue
|
||||
if len(examples) >= num_examples:
|
||||
break
|
||||
if key in seen_labels:
|
||||
continue
|
||||
seen_labels.add(key)
|
||||
t_tok, s_tok = per_text_tokens[idx]
|
||||
examples.append(
|
||||
{
|
||||
"text": text,
|
||||
"teacher_tokens": t_tok,
|
||||
"student_tokens": s_tok,
|
||||
"teacher_label": teacher_label,
|
||||
"student_pred": student_pred,
|
||||
"match": (student_pred == teacher_label) if student_pred else None,
|
||||
}
|
||||
)
|
||||
|
||||
return {
|
||||
"num_cases": n,
|
||||
"token_method": token_method,
|
||||
"fixed_prompt_overhead": fixed_overhead,
|
||||
"teacher_input_total": teacher_input_total,
|
||||
"teacher_input_avg": teacher_avg,
|
||||
"student_input_total": student_input_total,
|
||||
"student_input_avg": student_avg,
|
||||
"input_token_reduction_pct": reduction_pct,
|
||||
"teacher_student_ratio": ratio,
|
||||
"student_accuracy": accuracy,
|
||||
"student_correct": correct,
|
||||
"student_evaluated": evaluated,
|
||||
"examples": examples,
|
||||
}
|
||||
|
||||
|
||||
def print_report(r: Dict) -> None:
|
||||
line = "=" * 78
|
||||
print("\n" + line)
|
||||
print("Prompt 蒸馏：蒸馏前 vs 蒸馏后 量化对比")
|
||||
print(line)
|
||||
print(f"样本数 : {r['num_cases']}")
|
||||
print(f"Token 计数方式 : {r['token_method']}")
|
||||
print(f"固定提示开销 : {r['fixed_prompt_overhead']} tokens（模板本身，每次调用都要重复付费）")
|
||||
|
||||
print("\n" + "-" * 78)
|
||||
print("一、输入成本（每次调用的输入 token）")
|
||||
print("-" * 78)
|
||||
print(f"{'维度':<24}{'教师(长提示+思考)':>20}{'学生(无提示)':>18}")
|
||||
print(f"{'单条平均输入 token':<24}{r['teacher_input_avg']:>20.1f}{r['student_input_avg']:>18.1f}")
|
||||
print(f"{'全量总输入 token':<24}{r['teacher_input_total']:>20,}{r['student_input_total']:>18,}")
|
||||
print(
|
||||
f"\n→ 输入 token 降低 {r['input_token_reduction_pct']:.1f}%"
|
||||
f"（教师是学生的 {r['teacher_student_ratio']:.1f} 倍）。"
|
||||
)
|
||||
print(" 按输入 token 计费的 API 上，这一项直接等比例降低费用；教师端还有未计入的")
|
||||
print(" 思考（CoT）输出 token，实际差距只会更大。延迟需在 GPU 上实测，此处不估算。")
|
||||
|
||||
print("\n" + "-" * 78)
|
||||
print("二、任务质量（学生在相同输入上与教师标注的一致率 = 蒸馏保真度）")
|
||||
print("-" * 78)
|
||||
if r["student_accuracy"] is not None:
|
||||
print(
|
||||
f"教师(基准) : 100.00% 学生(蒸馏后) : {r['student_accuracy'] * 100:.2f}%"
|
||||
f" ({r['student_correct']}/{r['student_evaluated']})"
|
||||
)
|
||||
print(
|
||||
f"→ 无提示、无思考的学生保留了教师约 {r['student_accuracy'] * 100:.1f}% 的判断，"
|
||||
f"质量损失约 {(1 - r['student_accuracy']) * 100:.1f} 个百分点。"
|
||||
)
|
||||
else:
|
||||
print("未找到 evaluation_results.json（学生尚未评估）。先运行 evaluate.py 生成，")
|
||||
print("再回来看这一栏。本栏缺失不影响上面的输入成本对比。")
|
||||
|
||||
print("\n" + "-" * 78)
|
||||
print(f"三、逐条案例（{len(r['examples'])} 例）")
|
||||
print("-" * 78)
|
||||
print(f"{'待分类文本':<44}{'教师tok':>8}{'学生tok':>8}{'教师':>6}{'学生':>6}{'一致':>6}")
|
||||
for ex in r["examples"]:
|
||||
if ex["match"] is None:
|
||||
mark = "—"
|
||||
else:
|
||||
mark = "✓" if ex["match"] else "✗"
|
||||
pred = ex["student_pred"] if ex["student_pred"] else "—"
|
||||
print(
|
||||
f"{truncate(ex['text']):<44}"
|
||||
f"{ex['teacher_tokens']:>8}{ex['student_tokens']:>8}"
|
||||
f"{ex['teacher_label']:>6}{pred:>6}{mark:>6}"
|
||||
)
|
||||
print(line + "\n")
|
||||
|
||||
|
||||
def main():
|
||||
parser = argparse.ArgumentParser(
|
||||
description="Prompt 蒸馏「蒸馏前 vs 蒸馏后」量化对比：离线算出输入成本、任务质量与逐条案例",
|
||||
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
|
||||
)
|
||||
parser.add_argument(
|
||||
"--test_file",
|
||||
type=str,
|
||||
default="./example-data/multilingual.txt",
|
||||
help="待分类文本文件（每行一句），作为对比的输入集合",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--train_data_file",
|
||||
type=str,
|
||||
default="./data/prompt_distillation_lang.jsonl",
|
||||
help="蒸馏训练数据（教师标注），用于取教师标签作为质量基准",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--eval_results",
|
||||
type=str,
|
||||
default="./evaluation_results.json",
|
||||
help="evaluate.py 产出的评估结果，用于读取学生的一致率（可选）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--prompt_source",
|
||||
type=str,
|
||||
default="./create_data.py",
|
||||
help="包含教师提示模板 LANGUAGE_CLASSIFICATION_PROMPT 的源文件",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--tokenizer",
|
||||
type=str,
|
||||
default=None,
|
||||
help="可选：HuggingFace 分词器名/路径（如 Qwen/Qwen3-30B-A3B-Instruct-2507）。"
|
||||
"指定后用它精确计数；不指定则用 tiktoken 近似，保证离线可跑",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--num_examples",
|
||||
type=int,
|
||||
default=10,
|
||||
help="逐条案例展示的条数（尽量覆盖不同语言）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output_file",
|
||||
type=str,
|
||||
default=None,
|
||||
help="可选：把对比结果（含逐条案例）保存为 JSON 的路径",
|
||||
)
|
||||
args = parser.parse_args()
|
||||
|
||||
if not os.path.exists(args.test_file):
|
||||
raise FileNotFoundError(f"待分类文本文件不存在: {args.test_file}")
|
||||
if not os.path.exists(args.prompt_source):
|
||||
raise FileNotFoundError(f"提示模板源文件不存在: {args.prompt_source}")
|
||||
|
||||
prompt_template = load_prompt_template(args.prompt_source)
|
||||
texts = load_texts(args.test_file)
|
||||
teacher_labels = load_teacher_labels(args.train_data_file)
|
||||
eval_results = load_eval_results(args.eval_results)
|
||||
count_tokens, token_method = build_token_counter(args.tokenizer)
|
||||
|
||||
if not teacher_labels:
|
||||
print(
|
||||
f"[warn] 未从 {args.train_data_file} 读到教师标注，逐条案例的教师标签将显示为 '?'。",
|
||||
file=sys.stderr,
|
||||
)
|
||||
if eval_results is None:
|
||||
print(
|
||||
f"[warn] 未找到 {args.eval_results}，将只给出输入成本对比，跳过质量一栏。",
|
||||
file=sys.stderr,
|
||||
)
|
||||
|
||||
report = compare(
|
||||
prompt_template=prompt_template,
|
||||
texts=texts,
|
||||
teacher_labels=teacher_labels,
|
||||
eval_results=eval_results,
|
||||
count_tokens=count_tokens,
|
||||
token_method=token_method,
|
||||
num_examples=args.num_examples,
|
||||
)
|
||||
|
||||
print_report(report)
|
||||
|
||||
if args.output_file:
|
||||
with open(args.output_file, "w", encoding="utf-8") as f:
|
||||
json.dump(report, f, indent=2, ensure_ascii=False)
|
||||
print(f"📁 对比结果已保存到: {args.output_file}")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
@@ -0,0 +1,445 @@
|
||||
"""
|
||||
Data generation script for prompt distillation using vLLM.
|
||||
|
||||
This script generates training data for prompt distillation by using a teacher model
|
||||
to generate language classification labels with a detailed prompt, which will then be
|
||||
used to train a student model that internalizes the prompt.
|
||||
|
||||
Based on the tinker cookbook prompt distillation recipe.
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import asyncio
|
||||
import json
|
||||
import os
|
||||
import re
|
||||
from pathlib import Path
|
||||
from typing import Optional
|
||||
|
||||
from tqdm.asyncio import tqdm_asyncio
|
||||
|
||||
# 注意：vllm / SamplingParams 在 generate_distillation_data() 内部按需导入，
|
||||
# 这样即便未安装 vllm（如离线查看 --help 时）也能正常展示命令行帮助。
|
||||
|
||||
LANGUAGE_CLASSIFICATION_PROMPT = """You are a precise language classifier.
|
||||
|
||||
Goal: Classify the language of the provided text into exactly one of these labels:
|
||||
ar (Arabic), de (German), el (Greek), en (English), es (Spanish), fr (French),
|
||||
hi (Hindi), ru (Russian), tr (Turkish), ur (Urdu), vi (Vietnamese),
|
||||
zh (Chinese - Simplified), ot (Other/Unknown).
|
||||
|
||||
Instructions:
|
||||
1) Preprocess carefully (without changing the intended meaning):
|
||||
- Trim whitespace.
|
||||
- Ignore URLs, emails, file paths, hashtags, user handles, and emojis.
|
||||
- Ignore numbers, math expressions, and standalone punctuation.
|
||||
- If there is code, IGNORE code syntax (keywords, operators, braces) and focus ONLY on human language in comments and string literals.
|
||||
- Preserve letters and diacritics; do NOT strip accents.
|
||||
- If after ignoring the above there are no alphabetic letters left, output 'ot'.
|
||||
|
||||
2) Script-based rules (highest priority):
|
||||
- Devanagari script → hi.
|
||||
- Greek script → el.
|
||||
- Cyrillic script → ru.
|
||||
- Han characters (中文) → zh. (Treat Traditional as zh too.)
|
||||
- Arabic script → ar vs ur:
|
||||
• If Urdu-only letters appear (e.g., ے, ڑ, ں, ھ, ٹ, ڈ, کھ, گ, چ with Urdu forms), or clear Urdu words, choose ur.
|
||||
• Otherwise choose ar.
|
||||
(If multiple scripts appear, pick the script that contributes the majority of alphabetic characters. If tied, go to step 5.)
|
||||
|
||||
3) Latin-script heuristics (use when text is mainly Latin letters):
|
||||
- vi: presence of Vietnamese-specific letters/diacritics (ă â ê ô ơ ư đ, plus dense diacritics across many words).
|
||||
- tr: presence of Turkish-specific letters (ı İ ğ Ğ ş Ş ç Ç ö Ö ü Ü) and common function words (ve, bir, için, değil, ama, çok).
|
||||
- de: presence of umlauts (ä ö ü) or ß and common function words (und, der, die, das, nicht, ist).
|
||||
- es: presence of ñ, ¿, ¡ and common words (y, de, la, el, es, no, por, para, con, gracias, hola).
|
||||
- fr: frequent French diacritics (é è ê à ç ô â î û ù) and common words (et, le, la, les, des, une, est, avec, pour, merci, bonjour).
|
||||
- en: default among Latin languages if strong evidence for others is absent, but ONLY if English function words are present (the, and, is, are, to, of, in, for, on, with). If evidence is insufficient for any Latin language, prefer 'ot' over guessing.
|
||||
|
||||
4) Named entities & loanwords:
|
||||
- Do NOT decide based on a single proper noun, brand, or place name.
|
||||
- Require at least two function words or repeated language-specific signals (diacritics/letters) before assigning a Latin-language label.
|
||||
|
||||
5) Mixed-language text:
|
||||
- Determine the dominant language by counting indicative tokens (language-specific letters/diacritics/function words) AFTER preprocessing.
|
||||
- If two or more languages are equally dominant or the text is a deliberate multi-language mix, return 'ot'.
|
||||
|
||||
6) Very short or noisy inputs:
|
||||
- If the text is ≤2 meaningful words or too short to be confident, return 'ot' unless there is a very strong language-specific signal (e.g., "bonjour" → fr, "hola" → es).
|
||||
|
||||
7) Transliteration/romanization:
|
||||
- If Hindi/Urdu/Arabic/Chinese/Russian/Greek is written purely in Latin letters (romanized) without clear, repeated language-specific cue words, return 'ot'. (Only classify as hi/ur/ar/zh/ru/el when native scripts or highly distinctive romanized patterns are clearly present.)
|
||||
|
||||
8) Code-heavy inputs:
|
||||
- If the text is mostly code with minimal or no natural-language comments/strings, return 'ot'.
|
||||
- If comments/strings clearly indicate a language per rules above, use that label.
|
||||
|
||||
9) Ambiguity & confidence:
|
||||
- When in doubt, choose 'ot' rather than guessing.
|
||||
|
||||
Text to classify:
|
||||
{text}
|
||||
|
||||
Output format:
|
||||
- Respond with EXACTLY one line: "Final Answer: xx"
|
||||
- Where xx ∈ {{ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh, ot}} and nothing else.
|
||||
"""
|
||||
|
||||
|
||||
def parse_final_answer(response: str, debug: bool = False) -> Optional[str]:
|
||||
"""
|
||||
Parse the final answer from the model response.
|
||||
For Thinking models, extract from <think>...</think> tags or after them.
|
||||
"""
|
||||
# For Thinking models, the response may have <think></think> tags
|
||||
# Remove thinking content and focus on the final answer
|
||||
response_stripped = response.strip()
|
||||
|
||||
# Remove <think>...</think> content if present
|
||||
response_cleaned = re.sub(r'<think>.*?</think>', '', response_stripped, flags=re.DOTALL)
|
||||
response_cleaned = response_cleaned.strip()
|
||||
|
||||
# Also try the original response
|
||||
candidates = [response_cleaned, response_stripped]
|
||||
|
||||
valid_labels = {'ar', 'de', 'el', 'en', 'es', 'fr', 'hi', 'ru', 'tr', 'ur', 'vi', 'zh', 'ot'}
|
||||
|
||||
# Try multiple patterns to extract language label
|
||||
patterns = [
|
||||
r"Final Answer:\s*(\w{2})", # Standard format
|
||||
r"Final Answer:\s*([a-z]{2})", # Lowercase only
|
||||
r"Answer:\s*(\w{2})", # Without "Final"
|
||||
r"Language:\s*(\w{2})", # "Language: xx"
|
||||
r"^([a-z]{2})$", # Just the label alone
|
||||
r"\b([a-z]{2})\b\s*$", # Label at the end with word boundary
|
||||
r"is:\s*(\w{2})", # "is: xx"
|
||||
r"→\s*(\w{2})", # "→ xx"
|
||||
]
|
||||
|
||||
for candidate in candidates:
|
||||
candidate_lower = candidate.lower()
|
||||
|
||||
# Try each pattern
|
||||
for pattern in patterns:
|
||||
match = re.search(pattern, candidate_lower, re.MULTILINE)
|
||||
if match:
|
||||
label = match.group(1)
|
||||
if label in valid_labels:
|
||||
if debug:
|
||||
print(f" [DEBUG] Matched pattern '{pattern}' -> '{label}'")
|
||||
return label
|
||||
|
||||
# Special case: check if the entire response is just a language code
|
||||
if len(candidate) <= 3 and candidate_lower in valid_labels:
|
||||
if debug:
|
||||
print(f" [DEBUG] Matched entire response as label -> '{candidate_lower}'")
|
||||
return candidate_lower
|
||||
|
||||
if debug:
|
||||
print(f" [DEBUG] No pattern matched.")
|
||||
print(f" [DEBUG] Response length: {len(response_stripped)}")
|
||||
print(f" [DEBUG] Cleaned response: '{response_cleaned[:300]}'")
|
||||
print(f" [DEBUG] Original response: '{response_stripped[:300]}'")
|
||||
|
||||
return None
|
||||
|
||||
|
||||
async def generate_distillation_data(
|
||||
input_file: str,
|
||||
output_file: str,
|
||||
model_name: str = "Qwen/Qwen3-30B-A3B-Thinking-2507",
|
||||
temperature: float = 0.15,
|
||||
max_tokens: int = 4096,
|
||||
tensor_parallel_size: int = 1,
|
||||
max_retries: int = 3,
|
||||
):
|
||||
"""
|
||||
Generate prompt distillation training data.
|
||||
|
||||
Args:
|
||||
input_file: Path to file containing sentences to classify (one per line)
|
||||
output_file: Path to save the generated training data (JSONL format)
|
||||
model_name: Teacher model to use for generating labels
|
||||
temperature: Sampling temperature
|
||||
max_tokens: Maximum tokens to generate
|
||||
tensor_parallel_size: Number of GPUs to use for tensor parallelism
|
||||
"""
|
||||
print(f"Loading input sentences from {input_file}")
|
||||
with open(input_file, "r", encoding="utf-8") as f:
|
||||
sentences = [line.strip() for line in f if line.strip()]
|
||||
|
||||
print(f"Loaded {len(sentences)} sentences")
|
||||
if not sentences:
|
||||
print("Input file has no sentences to process, skipping data generation.")
|
||||
return
|
||||
|
||||
from vllm import LLM, SamplingParams
|
||||
|
||||
# Initialize vLLM model
|
||||
print(f"Initializing teacher model: {model_name}")
|
||||
print(f"Using tensor parallelism across {tensor_parallel_size} GPU(s)")
|
||||
|
||||
# Get tokenizer to use proper chat template
|
||||
from transformers import AutoTokenizer
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
||||
|
||||
llm = LLM(
|
||||
model=model_name,
|
||||
tensor_parallel_size=tensor_parallel_size,
|
||||
trust_remote_code=True,
|
||||
gpu_memory_utilization=0.90, # Use 90% of GPU memory for better throughput
|
||||
max_model_len=32768, # Match training max length
|
||||
enable_prefix_caching=True, # Cache the system prompt
|
||||
)
|
||||
|
||||
# Set sampling parameters - use Qwen3 recommended settings
|
||||
# For Thinking models, we need to allow enough tokens for reasoning
|
||||
sampling_params = SamplingParams(
|
||||
temperature=temperature,
|
||||
max_tokens=max_tokens,
|
||||
top_p=0.8,
|
||||
top_k=20,
|
||||
# Don't use custom stop sequences - let model finish naturally
|
||||
skip_special_tokens=False, # Keep special tokens for thinking models
|
||||
)
|
||||
|
||||
# Initial generation
|
||||
print("Generating labels with teacher model...")
|
||||
results = {} # sentence -> (response, final_answer)
|
||||
failed_indices = []
|
||||
failed_examples = [] # Store examples for debugging
|
||||
|
||||
# Format prompts using proper chat template
|
||||
print("Formatting prompts with Qwen3 chat template...")
|
||||
formatted_prompts = []
|
||||
for sentence in sentences:
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": LANGUAGE_CLASSIFICATION_PROMPT.format(text=sentence)
|
||||
}
|
||||
]
|
||||
# Use tokenizer's chat template
|
||||
prompt_text = tokenizer.apply_chat_template(
|
||||
messages,
|
||||
tokenize=False,
|
||||
add_generation_prompt=True
|
||||
)
|
||||
formatted_prompts.append(prompt_text)
|
||||
|
||||
print(f"Sample formatted prompt:")
|
||||
print(formatted_prompts[0])
|
||||
|
||||
outputs = llm.generate(formatted_prompts, sampling_params)
|
||||
|
||||
for idx, (sentence, output) in enumerate(zip(sentences, outputs)):
|
||||
response = output.outputs[0].text
|
||||
# Enable debug mode for first few failures
|
||||
debug_mode = len(failed_examples) < 3
|
||||
final_answer = parse_final_answer(response, debug=debug_mode)
|
||||
|
||||
if final_answer:
|
||||
results[sentence] = (response, final_answer)
|
||||
else:
|
||||
failed_indices.append(idx)
|
||||
# Store first 10 failed examples for debugging
|
||||
if len(failed_examples) < 10:
|
||||
failed_examples.append({
|
||||
'sentence': sentence,
|
||||
'response': response,
|
||||
})
|
||||
|
||||
# results is keyed by sentence text, so len(results) counts UNIQUE
|
||||
# sentences; the JSONL below writes one row per sentence occurrence. Count
|
||||
# rows actually labeled so the reported rate matches the output when the
|
||||
# corpus contains duplicate lines (common in language-ID data).
|
||||
num_labeled = sum(1 for s in sentences if s in results)
|
||||
print(f"\nInitial generation: {num_labeled}/{len(sentences)} successful ({num_labeled/len(sentences)*100:.2f}%)")
|
||||
|
||||
# Show debugging info for failed samples
|
||||
if failed_examples:
|
||||
print(f"\n{'='*60}")
|
||||
print("DEBUGGING: Examples of FAILED responses")
|
||||
print(f"{'='*60}")
|
||||
for i, example in enumerate(failed_examples, 1):
|
||||
print(f"\nFailed Example {i}:")
|
||||
print(f" Input: {example['sentence']}")
|
||||
print(f" Response: {example['response']}")
|
||||
print(f" Parsed result: None")
|
||||
|
||||
# Show examples of successful responses
|
||||
if results:
|
||||
print(f"\n{'='*60}")
|
||||
print("DEBUGGING: Examples of SUCCESSFUL responses")
|
||||
print(f"{'='*60}")
|
||||
success_examples = list(results.items())[:3]
|
||||
for i, (sentence, (response, label)) in enumerate(success_examples, 1):
|
||||
print(f"\nSuccess Example {i}:")
|
||||
print(f" Input: {sentence}")
|
||||
print(f" Response: {response}")
|
||||
print(f" Parsed label: {label}")
|
||||
|
||||
# Retry failed generations up to max_retries times
|
||||
for retry in range(1, max_retries + 1):
|
||||
if not failed_indices:
|
||||
break
|
||||
|
||||
print(f"\nRetry {retry}/{max_retries}: Regenerating {len(failed_indices)} failed samples...")
|
||||
|
||||
# Prepare prompts for failed sentences
|
||||
retry_sentences = [sentences[idx] for idx in failed_indices]
|
||||
retry_formatted_prompts = []
|
||||
for s in retry_sentences:
|
||||
messages = [
|
||||
{
|
||||
"role": "user",
|
||||
"content": LANGUAGE_CLASSIFICATION_PROMPT.format(text=s)
|
||||
}
|
||||
]
|
||||
prompt_text = tokenizer.apply_chat_template(
|
||||
messages,
|
||||
tokenize=False,
|
||||
add_generation_prompt=True
|
||||
)
|
||||
retry_formatted_prompts.append(prompt_text)
|
||||
|
||||
# Generate with slightly higher temperature to encourage different outputs
|
||||
retry_params = SamplingParams(
|
||||
temperature=min(temperature * (1 + retry * 0.1), 0.5), # Gradually increase temp
|
||||
max_tokens=max_tokens,
|
||||
top_p=0.8,
|
||||
top_k=20,
|
||||
skip_special_tokens=False,
|
||||
)
|
||||
|
||||
retry_outputs = llm.generate(retry_formatted_prompts, retry_params)
|
||||
|
||||
# Track newly successful and still-failed indices
|
||||
new_failed_indices = []
|
||||
for idx, sentence, output in zip(failed_indices, retry_sentences, retry_outputs):
|
||||
response = output.outputs[0].text
|
||||
final_answer = parse_final_answer(response)
|
||||
|
||||
if final_answer:
|
||||
results[sentence] = (response, final_answer)
|
||||
else:
|
||||
new_failed_indices.append(idx)
|
||||
|
||||
newly_successful = len(failed_indices) - len(new_failed_indices)
|
||||
print(f" ✓ {newly_successful} more samples successful")
|
||||
num_labeled = sum(1 for s in sentences if s in results)
|
||||
print(f" Total successful: {num_labeled}/{len(sentences)} ({num_labeled/len(sentences)*100:.2f}%)")
|
||||
|
||||
failed_indices = new_failed_indices
|
||||
|
||||
# Save results
|
||||
print(f"\nSaving results to {output_file}...")
|
||||
with open(output_file, "w", encoding="utf-8") as f:
|
||||
for sentence in sentences:
|
||||
if sentence in results:
|
||||
_, final_answer = results[sentence]
|
||||
data = {
|
||||
"messages": [
|
||||
{
|
||||
"role": "user",
|
||||
"content": sentence,
|
||||
},
|
||||
{
|
||||
"role": "assistant",
|
||||
"content": final_answer,
|
||||
},
|
||||
]
|
||||
}
|
||||
f.write(json.dumps(data, ensure_ascii=False) + "\n")
|
||||
|
||||
# Final report
|
||||
print(f"\n{'='*60}")
|
||||
print("DATA GENERATION COMPLETE")
|
||||
print(f"{'='*60}")
|
||||
num_labeled = sum(1 for s in sentences if s in results)
|
||||
print(f"Total sentences: {len(sentences)}")
|
||||
print(f"Valid labels generated: {num_labeled}")
|
||||
print(f"Failed after {max_retries} retries: {len(failed_indices)}")
|
||||
print(f"Final success rate: {num_labeled/len(sentences)*100:.2f}%")
|
||||
print(f"Saved to: {output_file}")
|
||||
|
||||
if failed_indices:
|
||||
print(f"\n⚠️ Warning: {len(failed_indices)} sentences failed to generate valid labels")
|
||||
print("Consider inspecting these samples or adjusting the prompt/temperature")
|
||||
|
||||
|
||||
def main():
|
||||
parser = argparse.ArgumentParser(
|
||||
description="用教师模型（长提示 + 思考）生成 Prompt 蒸馏训练数据",
|
||||
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
|
||||
)
|
||||
parser.add_argument(
|
||||
"--input_file",
|
||||
type=str,
|
||||
default="./example-data/multilingual.txt",
|
||||
help="输入文本文件路径（每行一句待分类文本）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output_file",
|
||||
type=str,
|
||||
default="./data/prompt_distillation_lang.jsonl",
|
||||
help="生成的训练数据保存路径（JSONL 格式）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--model_name",
|
||||
type=str,
|
||||
default="Qwen/Qwen3-30B-A3B-Thinking-2507",
|
||||
help="教师模型名称（用思考型模型以获得更高准确率）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--temperature",
|
||||
type=float,
|
||||
default=0.15,
|
||||
help="采样温度（与 tinker 保持一致，取 0.15）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--max_tokens",
|
||||
type=int,
|
||||
default=4096,
|
||||
help="单条生成的最大 token 数",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--tensor_parallel_size",
|
||||
type=int,
|
||||
default=1,
|
||||
help="张量并行使用的 GPU 数（30B 模型在 H100 上建议 2-4）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--max_retries",
|
||||
type=int,
|
||||
default=3,
|
||||
help="失败样本的最大重试次数",
|
||||
)
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
# Create output directory if needed
|
||||
output_dir = os.path.dirname(args.output_file)
|
||||
if output_dir:
|
||||
os.makedirs(output_dir, exist_ok=True)
|
||||
|
||||
# Check if input file exists
|
||||
if not os.path.exists(args.input_file):
|
||||
raise FileNotFoundError(f"Input file not found: {args.input_file}")
|
||||
|
||||
# Generate data
|
||||
asyncio.run(
|
||||
generate_distillation_data(
|
||||
input_file=args.input_file,
|
||||
output_file=args.output_file,
|
||||
model_name=args.model_name,
|
||||
temperature=args.temperature,
|
||||
max_tokens=args.max_tokens,
|
||||
tensor_parallel_size=args.tensor_parallel_size,
|
||||
max_retries=args.max_retries,
|
||||
)
|
||||
)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
|
||||
+191
@@ -0,0 +1,191 @@
|
||||
#!/bin/bash
|
||||
# Parallel data generation script for H100x8 GPU setup
|
||||
# This uses ALL 8 GPUs by running 2 instances in parallel
|
||||
|
||||
set -e
|
||||
|
||||
echo "=========================================="
|
||||
echo "Parallel Data Generation for H100x8"
|
||||
echo "=========================================="
|
||||
echo ""
|
||||
|
||||
# Configuration
|
||||
INPUT_FILE=${1:-"./example-data/multilingual.txt"}
|
||||
OUTPUT_FILE=${2:-"./data/prompt_distillation_lang.jsonl"}
|
||||
MODEL_NAME="Qwen/Qwen3-30B-A3B-Thinking-2507"
|
||||
|
||||
echo "Configuration:"
|
||||
echo " Input file: $INPUT_FILE"
|
||||
echo " Output file: $OUTPUT_FILE"
|
||||
echo " Model: $MODEL_NAME"
|
||||
echo " Strategy: 2 parallel instances, each using TP=4"
|
||||
echo ""
|
||||
|
||||
if [ ! -f "$INPUT_FILE" ]; then
|
||||
echo "❌ Input file not found: $INPUT_FILE"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Check if output file exists
|
||||
if [ -f "$OUTPUT_FILE" ]; then
|
||||
echo "⚠️ Output file already exists: $OUTPUT_FILE"
|
||||
read -p "Do you want to overwrite it? (y/N): " -n 1 -r
|
||||
echo
|
||||
if [[ ! $REPLY =~ ^[Yy]$ ]]; then
|
||||
echo "Aborted."
|
||||
exit 0
|
||||
fi
|
||||
rm "$OUTPUT_FILE"
|
||||
fi
|
||||
|
||||
# Create temp directory for split files
|
||||
TEMP_DIR="./data/temp_$$"
|
||||
mkdir -p "$TEMP_DIR"
|
||||
|
||||
echo "Splitting dataset into 2 parts..."
|
||||
TOTAL_LINES=$(wc -l < "$INPUT_FILE")
|
||||
HALF_LINES=$((TOTAL_LINES / 2))
|
||||
|
||||
head -n $HALF_LINES "$INPUT_FILE" > "$TEMP_DIR/part1.txt"
|
||||
tail -n +$((HALF_LINES + 1)) "$INPUT_FILE" > "$TEMP_DIR/part2.txt"
|
||||
|
||||
PART1_LINES=$(wc -l < "$TEMP_DIR/part1.txt")
|
||||
PART2_LINES=$(wc -l < "$TEMP_DIR/part2.txt")
|
||||
|
||||
echo " Part 1: $PART1_LINES lines (GPU 0-3)"
|
||||
echo " Part 2: $PART2_LINES lines (GPU 4-7)"
|
||||
echo ""
|
||||
|
||||
# Setup signal handler to kill both processes on Ctrl+C
|
||||
cleanup() {
|
||||
echo ""
|
||||
echo "🛑 Caught interrupt signal (Ctrl+C)"
|
||||
echo "Killing both processes..."
|
||||
if [ ! -z "$PID1" ] && kill -0 $PID1 2>/dev/null; then
|
||||
echo " Killing Instance 1 (PID $PID1)..."
|
||||
kill -TERM $PID1 2>/dev/null || true
|
||||
fi
|
||||
if [ ! -z "$PID2" ] && kill -0 $PID2 2>/dev/null; then
|
||||
echo " Killing Instance 2 (PID $PID2)..."
|
||||
kill -TERM $PID2 2>/dev/null || true
|
||||
fi
|
||||
# Wait a moment for graceful shutdown
|
||||
sleep 2
|
||||
# Force kill if still running
|
||||
if [ ! -z "$PID1" ] && kill -0 $PID1 2>/dev/null; then
|
||||
echo " Force killing Instance 1..."
|
||||
kill -9 $PID1 2>/dev/null || true
|
||||
fi
|
||||
if [ ! -z "$PID2" ] && kill -0 $PID2 2>/dev/null; then
|
||||
echo " Force killing Instance 2..."
|
||||
kill -9 $PID2 2>/dev/null || true
|
||||
fi
|
||||
# Cleanup temp files
|
||||
echo " Cleaning up temporary files..."
|
||||
rm -rf "$TEMP_DIR"
|
||||
echo "✅ Cleanup complete"
|
||||
exit 130
|
||||
}
|
||||
|
||||
trap cleanup SIGINT SIGTERM
|
||||
|
||||
# Run both instances in parallel
|
||||
echo "Starting parallel data generation..."
|
||||
echo ""
|
||||
|
||||
# Instance 1: GPU 0-3
|
||||
echo "Starting Instance 1 on GPUs 0-3..."
|
||||
CUDA_VISIBLE_DEVICES=0,1,2,3 python create_data.py \
|
||||
--input_file "$TEMP_DIR/part1.txt" \
|
||||
--output_file "$TEMP_DIR/part1.jsonl" \
|
||||
--model_name "$MODEL_NAME" \
|
||||
--temperature 0.15 \
|
||||
--tensor_parallel_size 4 \
|
||||
--max_retries 3 \
|
||||
> "$TEMP_DIR/instance1.log" 2>&1 &
|
||||
PID1=$!
|
||||
|
||||
echo "Instance 1 is running (PID: $PID1)"
|
||||
|
||||
# Instance 2: GPU 4-7
|
||||
echo "Starting Instance 2 on GPUs 4-7..."
|
||||
CUDA_VISIBLE_DEVICES=4,5,6,7 python create_data.py \
|
||||
--input_file "$TEMP_DIR/part2.txt" \
|
||||
--output_file "$TEMP_DIR/part2.jsonl" \
|
||||
--model_name "$MODEL_NAME" \
|
||||
--temperature 0.15 \
|
||||
--tensor_parallel_size 4 \
|
||||
--max_retries 3 \
|
||||
> "$TEMP_DIR/instance2.log" 2>&1 &
|
||||
PID2=$!
|
||||
|
||||
echo "Instance 2 is running (PID: $PID2)"
|
||||
echo ""
|
||||
echo "Both instances are running. You can monitor GPU usage with:"
|
||||
echo " watch -n 1 nvidia-smi"
|
||||
echo "Instance output is logged to:"
|
||||
echo " $TEMP_DIR/instance1.log"
|
||||
echo " $TEMP_DIR/instance2.log"
|
||||
echo ""
|
||||
|
||||
# Wait for both to complete
|
||||
echo "Waiting for both instances to complete..."
|
||||
echo " Instance 1 (PID $PID1): GPU 0-3"
|
||||
echo " Instance 2 (PID $PID2): GPU 4-7"
|
||||
echo ""
|
||||
|
||||
# `|| STATUS=` keeps a non-zero child exit from killing the script under
|
||||
# set -e — otherwise everything below (logs, combined check, cleanup)
|
||||
# is unreachable and the sibling instance is orphaned.
|
||||
STATUS1=0; wait $PID1 || STATUS1=$?
|
||||
echo ""
|
||||
echo "Instance 1 completed with status: $STATUS1"
|
||||
if [ $STATUS1 -ne 0 ]; then
|
||||
echo "Instance 1 log:"
|
||||
cat "$TEMP_DIR/instance1.log" | tail -50
|
||||
fi
|
||||
|
||||
STATUS2=0; wait $PID2 || STATUS2=$?
|
||||
echo ""
|
||||
echo "Instance 2 completed with status: $STATUS2"
|
||||
if [ $STATUS2 -ne 0 ]; then
|
||||
echo "Instance 2 log:"
|
||||
cat "$TEMP_DIR/instance2.log" | tail -50
|
||||
fi
|
||||
|
||||
echo ""
|
||||
|
||||
# Check if both succeeded
|
||||
if [ $STATUS1 -ne 0 ] || [ $STATUS2 -ne 0 ]; then
|
||||
echo "❌ One or both instances failed!"
|
||||
echo " Instance 1 status: $STATUS1"
|
||||
echo " Instance 2 status: $STATUS2"
|
||||
rm -rf "$TEMP_DIR"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Combine results
|
||||
echo "Combining results..."
|
||||
cat "$TEMP_DIR/part1.jsonl" "$TEMP_DIR/part2.jsonl" > "$OUTPUT_FILE"
|
||||
|
||||
# Show statistics
|
||||
PART1_COUNT=$(wc -l < "$TEMP_DIR/part1.jsonl")
|
||||
PART2_COUNT=$(wc -l < "$TEMP_DIR/part2.jsonl")
|
||||
TOTAL_COUNT=$((PART1_COUNT + PART2_COUNT))
|
||||
|
||||
echo ""
|
||||
echo "=========================================="
|
||||
echo "✅ Parallel data generation complete!"
|
||||
echo "=========================================="
|
||||
echo "Part 1: $PART1_COUNT / $PART1_LINES samples ($(awk "BEGIN {printf \"%.2f\", $PART1_COUNT/$PART1_LINES*100}")%)"
|
||||
echo "Part 2: $PART2_COUNT / $PART2_LINES samples ($(awk "BEGIN {printf \"%.2f\", $PART2_COUNT/$PART2_LINES*100}")%)"
|
||||
echo "Total: $TOTAL_COUNT / $TOTAL_LINES samples ($(awk "BEGIN {printf \"%.2f\", $TOTAL_COUNT/$TOTAL_LINES*100}")%)"
|
||||
echo ""
|
||||
echo "Output: $OUTPUT_FILE"
|
||||
echo ""
|
||||
|
||||
# Cleanup
|
||||
rm -rf "$TEMP_DIR"
|
||||
|
||||
echo "All 8 GPUs were utilized! 🚀"
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
Binary file not shown.
@@ -0,0 +1,539 @@
|
||||
"""
|
||||
Evaluation script for the distilled prompt model.
|
||||
|
||||
This script evaluates the student model's performance on language classification
|
||||
without providing the detailed prompt.
|
||||
|
||||
The student model (Qwen3-30B-A3B-Instruct) has been distilled from the teacher
|
||||
(Qwen3-30B-A3B-Thinking) which used a 2000+ token prompt. After distillation,
|
||||
the student responds directly without needing the prompt or thinking process.
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import re
|
||||
from pathlib import Path
|
||||
from typing import List, Dict, Optional
|
||||
from collections import defaultdict
|
||||
|
||||
import torch
|
||||
import numpy as np
|
||||
from transformers import AutoTokenizer, AutoModelForCausalLM
|
||||
from peft import PeftModel
|
||||
from tqdm import tqdm
|
||||
|
||||
|
||||
def load_model(model_path: str, base_model: str = "Qwen/Qwen3-30B-A3B-Instruct-2507"):
|
||||
"""Load the fine-tuned model with LoRA adapters."""
|
||||
print(f"Loading base model: {base_model}")
|
||||
print(f"This may take a few minutes for the 30B model...")
|
||||
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
|
||||
model = AutoModelForCausalLM.from_pretrained(
|
||||
base_model,
|
||||
dtype=torch.bfloat16, # Use 'dtype' instead of deprecated 'torch_dtype'
|
||||
device_map="auto",
|
||||
trust_remote_code=True,
|
||||
)
|
||||
|
||||
print(f"Loading LoRA adapters from: {model_path}")
|
||||
model = PeftModel.from_pretrained(model, model_path)
|
||||
model.eval()
|
||||
|
||||
return model, tokenizer
|
||||
|
||||
def compute_parse_rate(predicted: int, total: int) -> float:
|
||||
"""Parseable prediction rate; empty total is 0.0 (not ZeroDivisionError)."""
|
||||
return predicted / total if total > 0 else 0.0
|
||||
|
||||
|
||||
def format_pred_label(pred_label: Optional[str]) -> str:
|
||||
"""Display token for progress lines when the model reply is unparseable."""
|
||||
return pred_label or "??"
|
||||
|
||||
def parse_language_label(response: str) -> Optional[str]:
|
||||
"""Extract the language label from model response."""
|
||||
# Try to match common patterns
|
||||
patterns = [
|
||||
r"^([a-z]{2})$", # Just "en", "fr", etc.
|
||||
r"^([a-z]{2})\s*$", # With trailing whitespace
|
||||
r"Final Answer:\s*([a-z]{2})", # With "Final Answer:" prefix
|
||||
r"Language:\s*([a-z]{2})", # With "Language:" prefix
|
||||
]
|
||||
|
||||
response = response.strip().lower()
|
||||
for pattern in patterns:
|
||||
# response is lower-cased above, so the mixed-case "Final Answer:" /
|
||||
# "Language:" patterns only match case-insensitively — without this
|
||||
# they are unreachable and such answers score as unparseable.
|
||||
match = re.search(pattern, response, re.IGNORECASE)
|
||||
if match:
|
||||
return match.group(1)
|
||||
|
||||
# If response is short and looks like a language code
|
||||
if len(response) == 2 and response.isalpha():
|
||||
return response
|
||||
|
||||
return None
|
||||
|
||||
def evaluate_model(
|
||||
model,
|
||||
tokenizer,
|
||||
test_sentences: List[str],
|
||||
ground_truth_labels: Optional[List[str]] = None,
|
||||
max_new_tokens: int = 10,
|
||||
temperature: float = 0.0,
|
||||
) -> Dict:
|
||||
"""Evaluate the distilled model on test sentences."""
|
||||
|
||||
predictions = []
|
||||
correct = 0
|
||||
total = 0
|
||||
|
||||
print("\nEvaluating model...")
|
||||
print("="*80)
|
||||
|
||||
for idx, sentence in enumerate(test_sentences):
|
||||
# Format as user message (no system prompt!)
|
||||
messages = [
|
||||
{"role": "user", "content": sentence}
|
||||
]
|
||||
|
||||
# Apply chat template
|
||||
input_text = tokenizer.apply_chat_template(
|
||||
messages,
|
||||
tokenize=False,
|
||||
add_generation_prompt=True,
|
||||
)
|
||||
|
||||
# Tokenize
|
||||
inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
|
||||
|
||||
# Generate
|
||||
with torch.no_grad():
|
||||
outputs = model.generate(
|
||||
**inputs,
|
||||
max_new_tokens=max_new_tokens,
|
||||
temperature=temperature if temperature > 0 else None,
|
||||
do_sample=temperature > 0,
|
||||
pad_token_id=tokenizer.pad_token_id or tokenizer.eos_token_id,
|
||||
)
|
||||
|
||||
# Decode
|
||||
response = tokenizer.decode(
|
||||
outputs[0][inputs.input_ids.shape[1]:],
|
||||
skip_special_tokens=True,
|
||||
)
|
||||
|
||||
# Parse prediction
|
||||
pred_label = parse_language_label(response)
|
||||
predictions.append(pred_label)
|
||||
|
||||
# Check correctness and show real-time progress. Rows without ground
|
||||
# truth (teacher never labeled the sentence) are excluded from the
|
||||
# accuracy denominator — counting them as wrong deflated the reported
|
||||
# fidelity by exactly the unmatched fraction.
|
||||
if ground_truth_labels and idx < len(ground_truth_labels) and ground_truth_labels[idx] is not None:
|
||||
gt_label = ground_truth_labels[idx]
|
||||
is_correct = pred_label == gt_label
|
||||
if is_correct:
|
||||
correct += 1
|
||||
total += 1
|
||||
|
||||
# Show every sample in real-time
|
||||
status = "✓" if is_correct else "✗"
|
||||
status_color = "✓" if is_correct else "✗"
|
||||
|
||||
# Truncate sentence for display
|
||||
display_sentence = sentence if len(sentence) <= 60 else sentence[:57] + "..."
|
||||
|
||||
print(f"{status_color} [{idx+1:4d}/{len(test_sentences)}] "
|
||||
f"Pred: {format_pred_label(pred_label):>2s} | GT: {gt_label:>2s} | "
|
||||
f"Acc: {correct}/{total} ({correct/total*100:5.1f}%) | "
|
||||
f"{display_sentence}")
|
||||
else:
|
||||
# No ground truth - just show prediction
|
||||
display_sentence = sentence if len(sentence) <= 60 else sentence[:57] + "..."
|
||||
print(f" [{idx+1:4d}/{len(test_sentences)}] "
|
||||
f"Pred: {format_pred_label(pred_label):>2s} | "
|
||||
f"{display_sentence}")
|
||||
|
||||
print("="*80)
|
||||
print(f"Evaluation completed: {len(test_sentences)} samples processed")
|
||||
|
||||
# Calculate metrics
|
||||
results = {
|
||||
"predictions": predictions,
|
||||
"total": len(test_sentences),
|
||||
"predicted": sum(1 for p in predictions if p is not None),
|
||||
"unparseable": sum(1 for p in predictions if p is None),
|
||||
}
|
||||
|
||||
if ground_truth_labels:
|
||||
results["accuracy"] = correct / total if total > 0 else 0.0
|
||||
results["correct"] = correct
|
||||
results["evaluated"] = total
|
||||
|
||||
# Build confusion matrix
|
||||
results["confusion_matrix"] = build_confusion_matrix(
|
||||
predictions, ground_truth_labels, test_sentences
|
||||
)
|
||||
|
||||
return results
|
||||
|
||||
def build_confusion_matrix(predictions: List[str], ground_truth: List[str],
|
||||
sentences: List[str]) -> Dict:
|
||||
"""
|
||||
Build confusion matrix and identify problematic languages.
|
||||
|
||||
Returns:
|
||||
Dict with confusion matrix, per-language stats, and error examples
|
||||
"""
|
||||
# Get all unique labels (rows without ground truth are excluded).
|
||||
# Pair each prediction with its ground truth, then surface unparseable
|
||||
# predictions (pred is None, counted as the "None" column below) as an
|
||||
# explicit label. Otherwise confusion_matrix_2d drops that column and a
|
||||
# true-label row's cells no longer sum to that language's total.
|
||||
labeled = [(p, g) for p, g in zip(predictions, ground_truth) if g is not None]
|
||||
all_labels = sorted(set(g for _, g in labeled) | set(p for p, _ in labeled if p))
|
||||
if any(p is None for p, _ in labeled):
|
||||
all_labels = all_labels + ["None"]
|
||||
|
||||
# Initialize confusion matrix
|
||||
confusion = defaultdict(lambda: defaultdict(int))
|
||||
per_language_stats = defaultdict(lambda: {"correct": 0, "total": 0, "errors": []})
|
||||
|
||||
# Build matrix
|
||||
for idx, (pred, gt, sentence) in enumerate(zip(predictions, ground_truth, sentences)):
|
||||
if gt is None:
|
||||
continue
|
||||
if pred is None:
|
||||
pred = "None"
|
||||
|
||||
confusion[gt][pred] += 1
|
||||
per_language_stats[gt]["total"] += 1
|
||||
|
||||
if pred == gt:
|
||||
per_language_stats[gt]["correct"] += 1
|
||||
else:
|
||||
# Store error examples
|
||||
if len(per_language_stats[gt]["errors"]) < 5: # Keep first 5 errors per language
|
||||
per_language_stats[gt]["errors"].append({
|
||||
"sentence": sentence,
|
||||
"predicted": pred,
|
||||
"ground_truth": gt,
|
||||
"index": idx,
|
||||
})
|
||||
|
||||
# Calculate per-language accuracy
|
||||
language_accuracy = {}
|
||||
for lang in all_labels:
|
||||
stats = per_language_stats[lang]
|
||||
if stats["total"] > 0:
|
||||
accuracy = stats["correct"] / stats["total"]
|
||||
language_accuracy[lang] = {
|
||||
"accuracy": accuracy,
|
||||
"correct": stats["correct"],
|
||||
"total": stats["total"],
|
||||
"errors": stats["errors"],
|
||||
}
|
||||
|
||||
# Sort languages by accuracy (worst first)
|
||||
sorted_languages = sorted(
|
||||
language_accuracy.items(),
|
||||
key=lambda x: x[1]["accuracy"]
|
||||
)
|
||||
|
||||
# Create 2D confusion matrix as a proper array
|
||||
confusion_matrix_2d = []
|
||||
for true_label in all_labels:
|
||||
row = []
|
||||
for pred_label in all_labels:
|
||||
count = confusion.get(true_label, {}).get(pred_label, 0)
|
||||
row.append(count)
|
||||
confusion_matrix_2d.append(row)
|
||||
|
||||
return {
|
||||
"confusion_matrix": {gt: dict(preds) for gt, preds in confusion.items()}, # Dict format
|
||||
"confusion_matrix_2d": confusion_matrix_2d, # 2D array format
|
||||
"all_labels": all_labels,
|
||||
"per_language_accuracy": language_accuracy,
|
||||
"worst_performing_languages": sorted_languages[:5], # Top 5 worst
|
||||
"all_languages_sorted": sorted_languages, # All languages sorted by accuracy
|
||||
}
|
||||
|
||||
def print_confusion_matrix(confusion_data: Dict):
|
||||
"""Pretty print confusion matrix and analysis."""
|
||||
print("\n" + "="*80)
|
||||
print("CONFUSION MATRIX")
|
||||
print("="*80)
|
||||
|
||||
# Get labels and matrix
|
||||
labels = confusion_data["all_labels"]
|
||||
confusion = confusion_data["confusion_matrix"]
|
||||
|
||||
# Print matrix header - use shorter labels for display
|
||||
print("\n ", end="")
|
||||
for label in labels:
|
||||
# Truncate long labels for display
|
||||
display_label = label if len(label) <= 3 else label[:3]
|
||||
print(f"{display_label:>4s}", end="")
|
||||
print(" | Total")
|
||||
print(" " + "-" * (len(labels) * 4 + 10))
|
||||
|
||||
# Print matrix rows
|
||||
for true_label in labels:
|
||||
# Truncate long labels for display
|
||||
display_label = true_label if len(true_label) <= 2 else true_label[:2]
|
||||
print(f"{display_label:>2s} | ", end="")
|
||||
row_total = sum(confusion.get(true_label, {}).values())
|
||||
|
||||
for pred_label in labels:
|
||||
count = confusion.get(true_label, {}).get(pred_label, 0)
|
||||
if count > 0:
|
||||
if true_label == pred_label:
|
||||
print(f"\033[92m{count:4d}\033[0m", end="") # Green for diagonal
|
||||
else:
|
||||
print(f"\033[91m{count:4d}\033[0m", end="") # Red for errors
|
||||
else:
|
||||
print(f" .", end="")
|
||||
print(f" | {row_total:4d}")
|
||||
|
||||
# Print per-language accuracy
|
||||
print(f"\n{'='*80}")
|
||||
print("PER-LANGUAGE ACCURACY")
|
||||
print("="*80)
|
||||
|
||||
lang_acc = confusion_data["per_language_accuracy"]
|
||||
sorted_langs = sorted(lang_acc.items(), key=lambda x: x[1]["accuracy"])
|
||||
|
||||
for lang, stats in sorted_langs:
|
||||
acc = stats["accuracy"] * 100
|
||||
symbol = "✓" if acc >= 90 else "⚠️" if acc >= 70 else "✗"
|
||||
print(f"{symbol} {lang:>2s}: {acc:5.1f}% ({stats['correct']:4d}/{stats['total']:4d})")
|
||||
|
||||
# Identify problematic languages
|
||||
print(f"\n{'='*80}")
|
||||
print("MOST PROBLEMATIC LANGUAGES (Top 5)")
|
||||
print("="*80)
|
||||
|
||||
worst_languages = confusion_data["worst_performing_languages"]
|
||||
for idx, (lang, stats) in enumerate(worst_languages, 1):
|
||||
acc = stats["accuracy"] * 100
|
||||
print(f"\n{idx}. Language: {lang} - Accuracy: {acc:.1f}% ({stats['correct']}/{stats['total']})")
|
||||
|
||||
if stats["errors"]:
|
||||
print(f" Error examples:")
|
||||
for err in stats["errors"][:3]: # Show first 3 errors
|
||||
pred_lang = err['predicted']
|
||||
sentence_preview = err['sentence'][:50] + "..." if len(err['sentence']) > 50 else err['sentence']
|
||||
print(f" - Predicted {pred_lang} (should be {lang}): {sentence_preview}")
|
||||
|
||||
print("\n" + "="*80)
|
||||
|
||||
def main():
|
||||
parser = argparse.ArgumentParser(
|
||||
description="评估蒸馏后的学生模型（无提示、直接作答）在语言分类上的表现",
|
||||
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
|
||||
)
|
||||
parser.add_argument(
|
||||
"--model_path",
|
||||
type=str,
|
||||
default="./models/prompt_distillation_trl",
|
||||
help="训练得到的 LoRA adapter 路径",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--base_model",
|
||||
type=str,
|
||||
default="Qwen/Qwen3-30B-A3B-Instruct-2507",
|
||||
help="学生基座模型名称（非思考型）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--test_file",
|
||||
type=str,
|
||||
default="./example-data/multilingual.txt",
|
||||
help="测试文本文件（每行一句）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--ground_truth_file",
|
||||
type=str,
|
||||
default=None,
|
||||
help="标准答案标签文件（可选，每行一个）。未提供时会尝试从训练数据中读取",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--train_data_file",
|
||||
type=str,
|
||||
default="./data/prompt_distillation_lang.jsonl",
|
||||
help="训练数据文件，未提供 ground_truth_file 时从中提取教师标签作为基准",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output_file",
|
||||
type=str,
|
||||
default="./evaluation_results.json",
|
||||
help="评估结果的保存路径",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--max_samples",
|
||||
type=int,
|
||||
default=None,
|
||||
help="最多评估的样本数（用于快速测试）",
|
||||
)
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
# Load model
|
||||
model, tokenizer = load_model(args.model_path, args.base_model)
|
||||
|
||||
# Load test data
|
||||
print(f"\nLoading test sentences from: {args.test_file}")
|
||||
with open(args.test_file, "r", encoding="utf-8") as f:
|
||||
test_sentences = [line.strip() for line in f if line.strip()]
|
||||
|
||||
# Load ground truth
|
||||
ground_truth = None
|
||||
if args.ground_truth_file:
|
||||
print(f"Loading ground truth from: {args.ground_truth_file}")
|
||||
with open(args.ground_truth_file, "r", encoding="utf-8") as f:
|
||||
ground_truth = [line.strip() for line in f if line.strip()]
|
||||
elif args.train_data_file and Path(args.train_data_file).exists():
|
||||
# Try to extract ground truth from training data
|
||||
print(f"Loading ground truth from training data: {args.train_data_file}")
|
||||
ground_truth = []
|
||||
sentence_to_label = {}
|
||||
|
||||
with open(args.train_data_file, "r", encoding="utf-8") as f:
|
||||
for line in f:
|
||||
if line.strip():
|
||||
data = json.loads(line)
|
||||
messages = data.get("messages", [])
|
||||
if len(messages) >= 2:
|
||||
user_content = messages[0].get("content", "")
|
||||
assistant_content = messages[1].get("content", "")
|
||||
sentence_to_label[user_content] = assistant_content
|
||||
|
||||
# Match test sentences to ground truth labels. None = no ground truth
|
||||
# for this sentence; such rows are skipped in accuracy and the
|
||||
# confusion matrix instead of being scored as wrong under a phantom
|
||||
# "?" language.
|
||||
for sentence in test_sentences:
|
||||
ground_truth.append(sentence_to_label.get(sentence))
|
||||
|
||||
matched = sum(1 for gt in ground_truth if gt is not None)
|
||||
print(f"Matched {matched}/{len(test_sentences)} test sentences to training data")
|
||||
|
||||
if matched == 0:
|
||||
print("⚠️ Warning: No ground truth matched. Results will not show accuracy.")
|
||||
ground_truth = None
|
||||
|
||||
# Limit samples if requested
|
||||
if args.max_samples:
|
||||
test_sentences = test_sentences[:args.max_samples]
|
||||
if ground_truth:
|
||||
ground_truth = ground_truth[:args.max_samples]
|
||||
print(f"Limiting evaluation to {args.max_samples} samples")
|
||||
|
||||
print(f"Total test sentences: {len(test_sentences)}")
|
||||
|
||||
# Evaluate
|
||||
results = evaluate_model(
|
||||
model=model,
|
||||
tokenizer=tokenizer,
|
||||
test_sentences=test_sentences,
|
||||
ground_truth_labels=ground_truth,
|
||||
)
|
||||
|
||||
# Print confusion matrix analysis
|
||||
if "confusion_matrix" in results:
|
||||
print_confusion_matrix(results["confusion_matrix"])
|
||||
|
||||
# Print summary
|
||||
print("\n" + "="*60)
|
||||
print("EVALUATION SUMMARY")
|
||||
print("="*60)
|
||||
print(f"Model: {args.base_model}")
|
||||
print(f"Adapter: {args.model_path}")
|
||||
print(f"\nPerformance:")
|
||||
print(f" Total samples: {results['total']}")
|
||||
print(f" Successfully predicted: {results['predicted']}")
|
||||
print(f" Unparseable responses: {results['unparseable']}")
|
||||
rate = compute_parse_rate(results["predicted"], results["total"])
|
||||
print(f" Parse rate: {rate * 100:.2f}%")
|
||||
|
||||
if "accuracy" in results:
|
||||
print(f"\n Overall Accuracy: {results['accuracy']*100:.2f}%")
|
||||
print(f" Correct: {results['correct']}/{results['evaluated']}")
|
||||
print(f"\n💡 The model responds directly without the 2000+ token prompt!")
|
||||
|
||||
# Save comprehensive results to JSON
|
||||
output = {
|
||||
"model_path": args.model_path,
|
||||
"base_model": args.base_model,
|
||||
"test_file": args.test_file,
|
||||
"train_data_file": args.train_data_file,
|
||||
"timestamp": str(Path(args.model_path).stat().st_mtime) if Path(args.model_path).exists() else None,
|
||||
"summary": {
|
||||
"total_samples": results["total"],
|
||||
"predicted": results["predicted"],
|
||||
"unparseable": results["unparseable"],
|
||||
"parse_rate": compute_parse_rate(results["predicted"], results["total"]),
|
||||
},
|
||||
"predictions": results["predictions"],
|
||||
}
|
||||
|
||||
# Add accuracy metrics if available
|
||||
if "accuracy" in results:
|
||||
output["summary"]["accuracy"] = results["accuracy"]
|
||||
output["summary"]["correct"] = results["correct"]
|
||||
output["summary"]["evaluated"] = results["evaluated"]
|
||||
|
||||
# Add confusion matrix and language analysis
|
||||
if "confusion_matrix" in results:
|
||||
cm_data = results["confusion_matrix"]
|
||||
output["confusion_matrix"] = {
|
||||
"matrix_dict": cm_data["confusion_matrix"], # Dict format for readability
|
||||
"matrix_2d": cm_data["confusion_matrix_2d"], # 2D array for analysis
|
||||
"labels": cm_data["all_labels"], # Label order for the 2D matrix
|
||||
}
|
||||
|
||||
# Save ALL languages with their full statistics
|
||||
output["all_languages_accuracy"] = {
|
||||
lang: {
|
||||
"accuracy": stats["accuracy"],
|
||||
"correct": stats["correct"],
|
||||
"total": stats["total"],
|
||||
"error_examples": stats["errors"],
|
||||
}
|
||||
for lang, stats in cm_data["all_languages_sorted"]
|
||||
}
|
||||
|
||||
# Also save per-language accuracy (same data, different format)
|
||||
output["per_language_accuracy"] = {
|
||||
lang: {
|
||||
"accuracy": stats["accuracy"],
|
||||
"correct": stats["correct"],
|
||||
"total": stats["total"],
|
||||
"error_examples": stats["errors"],
|
||||
}
|
||||
for lang, stats in cm_data["per_language_accuracy"].items()
|
||||
}
|
||||
|
||||
# Save top 5 worst for quick reference
|
||||
output["worst_performing_languages"] = [
|
||||
{
|
||||
"language": lang,
|
||||
"accuracy": stats["accuracy"],
|
||||
"correct": stats["correct"],
|
||||
"total": stats["total"],
|
||||
"error_examples": stats["errors"],
|
||||
}
|
||||
for lang, stats in cm_data["worst_performing_languages"]
|
||||
]
|
||||
|
||||
# Save to file
|
||||
with open(args.output_file, "w", encoding="utf-8") as f:
|
||||
json.dump(output, f, indent=2, ensure_ascii=False)
|
||||
print(f"\n📁 Complete results saved to: {args.output_file}")
|
||||
print(f" Includes: predictions, confusion matrix, per-language stats, error examples")
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
|
||||
File diff suppressed because it is too large
Load Diff
File diff suppressed because it is too large
Load Diff
@@ -0,0 +1,25 @@
|
||||
# Prompt Distillation Requirements
|
||||
#
|
||||
# Core dependencies for reproducing the tinker prompt distillation experiment
|
||||
|
||||
# vLLM for efficient inference (data generation with teacher model)
|
||||
vllm>=0.6.0
|
||||
|
||||
# Progress bars and utilities
|
||||
tqdm>=4.65.0
|
||||
numpy>=1.24.0
|
||||
|
||||
# Training framework - Hugging Face TRL (Transformers Reinforcement Learning)
|
||||
trl>=0.9.0
|
||||
peft>=0.11.0
|
||||
accelerate>=0.28.0
|
||||
transformers>=4.36.0
|
||||
datasets>=2.14.0
|
||||
torch>=2.0.0
|
||||
|
||||
# Logging and monitoring
|
||||
wandb>=0.15.0
|
||||
|
||||
# Note: TRL provides a simple, well-documented API for supervised fine-tuning
|
||||
# with LoRA and works directly with JSONL data format
|
||||
|
||||
@@ -0,0 +1,752 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Run the manuscript's Experiment 8-8 with canonical, leakage-free evidence.
|
||||
|
||||
The campaign deliberately uses a small student so that *real parameter training*
|
||||
can run on Apple Silicon. It still preserves the experiment's causal contrast:
|
||||
|
||||
* teacher: long task prompt + a thinking model (Moonshot ``kimi-k3``);
|
||||
* student: raw user text only + a non-thinking 135M model;
|
||||
* data: disjoint public train/test splits with independent gold labels;
|
||||
* evidence: provider response IDs/usage, adapter weights, hashes, held-out
|
||||
quality, actual wall-clock latency, tokenizer counts and provider cost.
|
||||
|
||||
No API key or credential value is ever written to disk.
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import argparse
|
||||
import asyncio
|
||||
import hashlib
|
||||
import json
|
||||
import math
|
||||
import os
|
||||
import random
|
||||
import re
|
||||
import statistics
|
||||
import sys
|
||||
import time
|
||||
from dataclasses import dataclass
|
||||
from datetime import datetime, timezone
|
||||
from pathlib import Path
|
||||
from typing import Any, Iterable, Optional
|
||||
|
||||
|
||||
EXPERIMENT_ID = "8-8"
|
||||
DATASET_ID = "papluca/language-identification"
|
||||
DATASET_REVISION = "aa56583bf2bc52b0565770607d6fc3faebecf9e2"
|
||||
STUDENT_MODEL = "HuggingFaceTB/SmolLM2-135M-Instruct"
|
||||
STUDENT_REVISION = "12fd25f77366fa6b3b4b768ec3050bf629380bac"
|
||||
TEACHER_MODEL = "kimi-k3"
|
||||
TEACHER_BASE_URL = "https://api.moonshot.cn/v1"
|
||||
TEACHER_KEY_ENV = "MOONSHOT_API_KEY"
|
||||
|
||||
# Dated native Kimi K3 pricing copied from the verified Chapter 6 campaign
|
||||
# configuration. We preserve both the native rate and the dated FX rate.
|
||||
PRICING = {
|
||||
"input_per_million": 20.0,
|
||||
"cached_input_per_million": 2.0,
|
||||
"output_per_million": 100.0,
|
||||
"currency": "CNY",
|
||||
"source_url": "https://platform.kimi.com/docs/pricing/chat-k3.md",
|
||||
"as_of": "2026-07-29",
|
||||
"usd_per_currency_unit": 0.1477922077922078,
|
||||
"fx_source_url": "https://www.ecb.europa.eu/stats/eurofxref/eurofxref-daily.xml",
|
||||
"fx_as_of": "2026-07-29",
|
||||
}
|
||||
|
||||
TARGET_LABELS = {"ar", "de", "el", "en", "es", "fr", "hi", "ru", "tr", "ur", "vi", "zh"}
|
||||
VALID_LABELS = TARGET_LABELS | {"ot"}
|
||||
|
||||
LANGUAGE_CLASSIFICATION_PROMPT = """You are a precise language classifier.
|
||||
|
||||
Goal: Classify the language of the provided text into exactly one label:
|
||||
ar (Arabic), de (German), el (Greek), en (English), es (Spanish), fr (French),
|
||||
hi (Hindi), ru (Russian), tr (Turkish), ur (Urdu), vi (Vietnamese),
|
||||
zh (Chinese), or ot (all other languages / unknown).
|
||||
|
||||
Rules:
|
||||
1. Ignore URLs, email addresses, numbers, emoji, punctuation, and code syntax.
|
||||
2. Use native script first. Distinguish Arabic from Urdu by Urdu-specific
|
||||
letters and words. Treat both Simplified and Traditional Chinese as zh.
|
||||
3. For Latin scripts use vocabulary, function words, and diacritics together;
|
||||
do not decide from one proper noun.
|
||||
4. Deliberate mixed-language text or a language outside the listed set is ot.
|
||||
5. Very short text is ot unless the language signal is unambiguous.
|
||||
6. Think through ambiguity before answering, but keep that reasoning private.
|
||||
|
||||
Text to classify:
|
||||
{text}
|
||||
|
||||
The visible response must be exactly one line: Final Answer: xx
|
||||
"""
|
||||
|
||||
|
||||
def utc_now() -> str:
|
||||
return datetime.now(timezone.utc).isoformat()
|
||||
|
||||
|
||||
def sha256_file(path: Path) -> str:
|
||||
h = hashlib.sha256()
|
||||
with path.open("rb") as f:
|
||||
for chunk in iter(lambda: f.read(1024 * 1024), b""):
|
||||
h.update(chunk)
|
||||
return h.hexdigest()
|
||||
|
||||
|
||||
def canonical_hash(value: Any) -> str:
|
||||
payload = json.dumps(value, sort_keys=True, ensure_ascii=False, separators=(",", ":"))
|
||||
return hashlib.sha256(payload.encode("utf-8")).hexdigest()
|
||||
|
||||
|
||||
def write_json(path: Path, value: Any) -> None:
|
||||
path.parent.mkdir(parents=True, exist_ok=True)
|
||||
path.write_text(json.dumps(value, indent=2, ensure_ascii=False) + "\n", encoding="utf-8")
|
||||
|
||||
|
||||
def write_jsonl(path: Path, rows: Iterable[dict[str, Any]]) -> None:
|
||||
path.parent.mkdir(parents=True, exist_ok=True)
|
||||
with path.open("w", encoding="utf-8") as f:
|
||||
for row in rows:
|
||||
f.write(json.dumps(row, ensure_ascii=False) + "\n")
|
||||
|
||||
|
||||
def read_jsonl(path: Path) -> list[dict[str, Any]]:
|
||||
if not path.exists():
|
||||
return []
|
||||
with path.open(encoding="utf-8") as f:
|
||||
return [json.loads(line) for line in f if line.strip()]
|
||||
|
||||
|
||||
def map_gold(source_label: str) -> str:
|
||||
return source_label if source_label in TARGET_LABELS else "ot"
|
||||
|
||||
|
||||
def _stable_sample(rows: list[dict[str, str]], n: int, seed: int, split: str) -> list[dict[str, str]]:
|
||||
grouped: dict[str, list[dict[str, str]]] = {}
|
||||
for row in rows:
|
||||
grouped.setdefault(row["labels"], []).append(row)
|
||||
selected: list[dict[str, str]] = []
|
||||
for source_label in sorted(grouped):
|
||||
candidates = sorted(
|
||||
grouped[source_label],
|
||||
key=lambda r: hashlib.sha256(f"{seed}:{split}:{source_label}:{r['text']}".encode()).hexdigest(),
|
||||
)
|
||||
for rank, row in enumerate(candidates[:n]):
|
||||
selected.append(
|
||||
{
|
||||
"id": f"{split}-{source_label}-{rank:03d}",
|
||||
"split": split,
|
||||
"source_label": source_label,
|
||||
"gold_label": map_gold(source_label),
|
||||
"text": row["text"],
|
||||
}
|
||||
)
|
||||
return sorted(selected, key=lambda r: r["id"])
|
||||
|
||||
|
||||
def prepare_benchmark(run_dir: Path, train_per_language: int, test_per_language: int, seed: int) -> dict[str, Any]:
|
||||
from datasets import load_dataset
|
||||
|
||||
train_ds = load_dataset(DATASET_ID, revision=DATASET_REVISION, split="train")
|
||||
test_ds = load_dataset(DATASET_ID, revision=DATASET_REVISION, split="test")
|
||||
train = _stable_sample(list(train_ds), train_per_language, seed, "train")
|
||||
test = _stable_sample(list(test_ds), test_per_language, seed, "test")
|
||||
train_texts = {r["text"] for r in train}
|
||||
test_texts = {r["text"] for r in test}
|
||||
overlap = train_texts & test_texts
|
||||
if overlap:
|
||||
raise RuntimeError(f"train/test leakage detected: {len(overlap)} texts")
|
||||
write_jsonl(run_dir / "benchmark_train_gold.jsonl", train)
|
||||
write_jsonl(run_dir / "benchmark_test_gold.jsonl", test)
|
||||
provenance = {
|
||||
"dataset": DATASET_ID,
|
||||
"revision": DATASET_REVISION,
|
||||
"train_source_split": "train",
|
||||
"test_source_split": "test",
|
||||
"train_fingerprint": train_ds._fingerprint,
|
||||
"test_fingerprint": test_ds._fingerprint,
|
||||
"seed": seed,
|
||||
"train_per_source_language": train_per_language,
|
||||
"test_per_source_language": test_per_language,
|
||||
"train_rows": len(train),
|
||||
"test_rows": len(test),
|
||||
"train_unique_texts": len(train_texts),
|
||||
"test_unique_texts": len(test_texts),
|
||||
"exact_text_overlap": len(overlap),
|
||||
"train_rows_sha256": canonical_hash(train),
|
||||
"test_rows_sha256": canonical_hash(test),
|
||||
}
|
||||
write_json(run_dir / "dataset_provenance.json", provenance)
|
||||
return provenance
|
||||
|
||||
|
||||
def parse_label(text: Optional[str]) -> Optional[str]:
|
||||
if not text:
|
||||
return None
|
||||
matches = re.findall(r"Final Answer:\s*([a-z]{2})", text, re.IGNORECASE)
|
||||
if matches and matches[-1].lower() in VALID_LABELS:
|
||||
return matches[-1].lower()
|
||||
stripped = text.strip().lower()
|
||||
return stripped if stripped in VALID_LABELS else None
|
||||
|
||||
|
||||
def usage_cost(usage: dict[str, Any]) -> dict[str, float]:
|
||||
prompt = int(usage.get("prompt_tokens") or 0)
|
||||
completion = int(usage.get("completion_tokens") or 0)
|
||||
details = usage.get("prompt_tokens_details") or {}
|
||||
cached = int(details.get("cached_tokens") or usage.get("cached_tokens") or 0)
|
||||
uncached = max(0, prompt - cached)
|
||||
cny = (
|
||||
uncached * PRICING["input_per_million"]
|
||||
+ cached * PRICING["cached_input_per_million"]
|
||||
+ completion * PRICING["output_per_million"]
|
||||
) / 1_000_000
|
||||
return {"cny": cny, "usd": cny * PRICING["usd_per_currency_unit"]}
|
||||
|
||||
|
||||
async def collect_teacher(run_dir: Path, concurrency: int, max_retries: int) -> dict[str, Any]:
|
||||
from openai import AsyncOpenAI
|
||||
|
||||
key = os.getenv(TEACHER_KEY_ENV)
|
||||
if not key:
|
||||
raise RuntimeError(f"{TEACHER_KEY_ENV} is not configured")
|
||||
train = read_jsonl(run_dir / "benchmark_train_gold.jsonl")
|
||||
test = read_jsonl(run_dir / "benchmark_test_gold.jsonl")
|
||||
rows = train + test
|
||||
raw_path = run_dir / "teacher_receipts.jsonl"
|
||||
existing = {r["id"]: r for r in read_jsonl(raw_path)}
|
||||
client = AsyncOpenAI(api_key=key, base_url=TEACHER_BASE_URL, timeout=90)
|
||||
sem = asyncio.Semaphore(concurrency)
|
||||
lock = asyncio.Lock()
|
||||
|
||||
async def one(row: dict[str, Any]) -> dict[str, Any]:
|
||||
async with sem:
|
||||
error: Optional[str] = None
|
||||
for attempt in range(max_retries + 1):
|
||||
started = utc_now()
|
||||
t0 = time.perf_counter()
|
||||
try:
|
||||
response = await client.chat.completions.create(
|
||||
model=TEACHER_MODEL,
|
||||
messages=[
|
||||
{
|
||||
"role": "system",
|
||||
"content": "Reason privately. The visible response must be exactly: Final Answer: xx",
|
||||
},
|
||||
{"role": "user", "content": LANGUAGE_CLASSIFICATION_PROMPT.format(text=row["text"])},
|
||||
],
|
||||
max_tokens=256,
|
||||
extra_body={"reasoning_effort": "low"},
|
||||
)
|
||||
latency = time.perf_counter() - t0
|
||||
message = response.choices[0].message
|
||||
content = message.content or ""
|
||||
reasoning = getattr(message, "reasoning_content", None) or getattr(message, "reasoning", None)
|
||||
usage = response.usage.model_dump() if response.usage else {}
|
||||
record = {
|
||||
**row,
|
||||
"provider": "moonshot",
|
||||
"base_url": TEACHER_BASE_URL,
|
||||
"model_requested": TEACHER_MODEL,
|
||||
"response_id": response.id,
|
||||
"response_model": response.model,
|
||||
"response_created": response.created,
|
||||
"request_started_at": started,
|
||||
"latency_seconds": latency,
|
||||
"attempt": attempt,
|
||||
"content": content,
|
||||
"reasoning_content": reasoning,
|
||||
"prediction": parse_label(content),
|
||||
"usage": usage,
|
||||
"calculated_cost": usage_cost(usage),
|
||||
"request": {
|
||||
"max_tokens": 256,
|
||||
"reasoning_effort": "low",
|
||||
"prompt_sha256": hashlib.sha256(
|
||||
LANGUAGE_CLASSIFICATION_PROMPT.format(text=row["text"]).encode("utf-8")
|
||||
).hexdigest(),
|
||||
},
|
||||
"error": None,
|
||||
}
|
||||
async with lock:
|
||||
with raw_path.open("a", encoding="utf-8") as f:
|
||||
f.write(json.dumps(record, ensure_ascii=False) + "\n")
|
||||
return record
|
||||
except Exception as exc: # noqa: BLE001
|
||||
error = f"{type(exc).__name__}: {exc}"
|
||||
record = {**row, "provider": "moonshot", "model_requested": TEACHER_MODEL, "error": error}
|
||||
async with lock:
|
||||
with raw_path.open("a", encoding="utf-8") as f:
|
||||
f.write(json.dumps(record, ensure_ascii=False) + "\n")
|
||||
return record
|
||||
|
||||
pending = [row for row in rows if row["id"] not in existing or existing[row["id"]].get("error")]
|
||||
if pending:
|
||||
results = await asyncio.gather(*(one(row) for row in pending))
|
||||
existing.update({r["id"]: r for r in results})
|
||||
ordered = [existing[row["id"]] for row in rows]
|
||||
write_jsonl(raw_path, ordered)
|
||||
|
||||
accepted_train = [
|
||||
{
|
||||
"id": r["id"],
|
||||
"messages": [
|
||||
{"role": "user", "content": r["text"]},
|
||||
{"role": "assistant", "content": r["prediction"]},
|
||||
],
|
||||
"teacher_response_id": r.get("response_id"),
|
||||
"gold_label": r["gold_label"],
|
||||
}
|
||||
for r in ordered
|
||||
if r["split"] == "train" and not r.get("error") and r.get("prediction") == r["gold_label"]
|
||||
]
|
||||
write_jsonl(run_dir / "student_train.jsonl", accepted_train)
|
||||
|
||||
def split_summary(split: str) -> dict[str, Any]:
|
||||
rs = [r for r in ordered if r["split"] == split]
|
||||
valid = [r for r in rs if not r.get("error") and r.get("prediction")]
|
||||
costs = [r.get("calculated_cost") or {} for r in valid]
|
||||
return {
|
||||
"rows": len(rs),
|
||||
"valid_receipts": len(valid),
|
||||
"unique_response_ids": len({r.get("response_id") for r in valid}),
|
||||
"correct": sum(r.get("prediction") == r["gold_label"] for r in valid),
|
||||
"gold_accuracy": sum(r.get("prediction") == r["gold_label"] for r in valid) / len(rs),
|
||||
"prompt_tokens": sum((r.get("usage") or {}).get("prompt_tokens", 0) for r in valid),
|
||||
"completion_tokens": sum((r.get("usage") or {}).get("completion_tokens", 0) for r in valid),
|
||||
"latency_seconds_total": sum(r.get("latency_seconds", 0) for r in valid),
|
||||
"latency_seconds_mean": statistics.mean(r.get("latency_seconds", 0) for r in valid) if valid else None,
|
||||
"provider_cost_cny": sum(c.get("cny", 0) for c in costs),
|
||||
"provider_cost_usd": sum(c.get("usd", 0) for c in costs),
|
||||
}
|
||||
|
||||
summary = {
|
||||
"teacher": {"provider": "moonshot", "model": TEACHER_MODEL, "pricing": PRICING},
|
||||
"train": split_summary("train"),
|
||||
"test": split_summary("test"),
|
||||
"accepted_student_train_rows": len(accepted_train),
|
||||
}
|
||||
write_json(run_dir / "teacher_summary.json", summary)
|
||||
return summary
|
||||
|
||||
|
||||
@dataclass
|
||||
class EncodedRow:
|
||||
input_ids: list[int]
|
||||
labels: list[int]
|
||||
|
||||
|
||||
def _chat_template_ids(encoded: Any) -> list[int]:
|
||||
"""Return one token-id list across Transformers 4.x and 5.x APIs."""
|
||||
if isinstance(encoded, dict) or hasattr(encoded, "keys"):
|
||||
encoded = encoded["input_ids"]
|
||||
if hasattr(encoded, "tolist"):
|
||||
encoded = encoded.tolist()
|
||||
if encoded and isinstance(encoded[0], list):
|
||||
if len(encoded) != 1:
|
||||
raise ValueError("expected one chat-template sequence")
|
||||
encoded = encoded[0]
|
||||
if not isinstance(encoded, list) or not all(isinstance(token, int) for token in encoded):
|
||||
raise TypeError("chat template did not return a one-dimensional integer token sequence")
|
||||
return encoded
|
||||
|
||||
|
||||
def _local_device(torch: Any) -> Any:
|
||||
if torch.cuda.is_available():
|
||||
return torch.device("cuda")
|
||||
if torch.backends.mps.is_available():
|
||||
return torch.device("mps")
|
||||
return torch.device("cpu")
|
||||
|
||||
|
||||
def _encode_sft(tokenizer: Any, messages: list[dict[str, str]], max_length: int) -> EncodedRow:
|
||||
prompt = messages[:-1]
|
||||
prompt_ids = _chat_template_ids(
|
||||
tokenizer.apply_chat_template(prompt, tokenize=True, add_generation_prompt=True)
|
||||
)
|
||||
full_ids = _chat_template_ids(
|
||||
tokenizer.apply_chat_template(messages, tokenize=True, add_generation_prompt=False)
|
||||
)
|
||||
full_ids = full_ids[:max_length]
|
||||
labels = [-100] * min(len(prompt_ids), len(full_ids)) + full_ids[min(len(prompt_ids), len(full_ids)) :]
|
||||
return EncodedRow(input_ids=full_ids, labels=labels)
|
||||
|
||||
|
||||
def train_student(run_dir: Path, epochs: int, batch_size: int, learning_rate: float, seed: int) -> dict[str, Any]:
|
||||
import torch
|
||||
from peft import LoraConfig, get_peft_model
|
||||
from torch.utils.data import DataLoader
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
|
||||
random.seed(seed)
|
||||
torch.manual_seed(seed)
|
||||
device = _local_device(torch)
|
||||
tokenizer = AutoTokenizer.from_pretrained(STUDENT_MODEL, revision=STUDENT_REVISION)
|
||||
if tokenizer.pad_token_id is None:
|
||||
tokenizer.pad_token = tokenizer.eos_token
|
||||
rows = read_jsonl(run_dir / "student_train.jsonl")
|
||||
if not rows:
|
||||
# A teacher campaign may be interrupted after individual receipts have
|
||||
# been durably appended but before its final derived files are written.
|
||||
# Rebuild the training view from successful, gold-verified real
|
||||
# receipts so the documented phase-resume boundary is actually usable.
|
||||
receipts = read_jsonl(run_dir / "teacher_receipts.jsonl")
|
||||
rows = [
|
||||
{
|
||||
"id": receipt["id"],
|
||||
"messages": [
|
||||
{"role": "user", "content": receipt["text"]},
|
||||
{"role": "assistant", "content": receipt["prediction"]},
|
||||
],
|
||||
"teacher_response_id": receipt.get("response_id"),
|
||||
"gold_label": receipt["gold_label"],
|
||||
}
|
||||
for receipt in receipts
|
||||
if receipt.get("split") == "train"
|
||||
and not receipt.get("error")
|
||||
and receipt.get("response_id")
|
||||
and receipt.get("prediction") == receipt.get("gold_label")
|
||||
]
|
||||
if rows:
|
||||
write_jsonl(run_dir / "student_train.jsonl", rows)
|
||||
if not rows:
|
||||
raise RuntimeError("no accepted teacher rows available for training")
|
||||
encoded = [_encode_sft(tokenizer, r["messages"], 192) for r in rows]
|
||||
|
||||
def collate(items: list[EncodedRow]) -> dict[str, torch.Tensor]:
|
||||
length = max(len(x.input_ids) for x in items)
|
||||
input_ids, labels, masks = [], [], []
|
||||
for item in items:
|
||||
pad = length - len(item.input_ids)
|
||||
input_ids.append(item.input_ids + [tokenizer.pad_token_id] * pad)
|
||||
labels.append(item.labels + [-100] * pad)
|
||||
masks.append([1] * len(item.input_ids) + [0] * pad)
|
||||
return {
|
||||
"input_ids": torch.tensor(input_ids, dtype=torch.long),
|
||||
"labels": torch.tensor(labels, dtype=torch.long),
|
||||
"attention_mask": torch.tensor(masks, dtype=torch.long),
|
||||
}
|
||||
|
||||
generator = torch.Generator().manual_seed(seed)
|
||||
loader = DataLoader(encoded, batch_size=batch_size, shuffle=True, collate_fn=collate, generator=generator)
|
||||
base = AutoModelForCausalLM.from_pretrained(
|
||||
STUDENT_MODEL,
|
||||
revision=STUDENT_REVISION,
|
||||
dtype=torch.bfloat16 if device.type == "cuda" else torch.float32,
|
||||
)
|
||||
config = LoraConfig(
|
||||
r=16,
|
||||
lora_alpha=32,
|
||||
lora_dropout=0.0,
|
||||
target_modules=["q_proj", "v_proj"],
|
||||
bias="none",
|
||||
task_type="CAUSAL_LM",
|
||||
)
|
||||
model = get_peft_model(base, config).to(device)
|
||||
model.train()
|
||||
params = [p for p in model.parameters() if p.requires_grad]
|
||||
optimizer = torch.optim.AdamW(params, lr=learning_rate)
|
||||
losses: list[float] = []
|
||||
started = utc_now()
|
||||
t0 = time.perf_counter()
|
||||
for epoch in range(epochs):
|
||||
for batch in loader:
|
||||
optimizer.zero_grad(set_to_none=True)
|
||||
batch = {k: v.to(device) for k, v in batch.items()}
|
||||
output = model(**batch)
|
||||
output.loss.backward()
|
||||
torch.nn.utils.clip_grad_norm_(params, 1.0)
|
||||
optimizer.step()
|
||||
losses.append(float(output.loss.detach().cpu()))
|
||||
if device.type == "mps":
|
||||
torch.mps.synchronize()
|
||||
runtime = time.perf_counter() - t0
|
||||
adapter_dir = run_dir / "student_adapter"
|
||||
adapter_dir.mkdir(parents=True, exist_ok=True)
|
||||
model.save_pretrained(adapter_dir, safe_serialization=True)
|
||||
tokenizer.save_pretrained(adapter_dir)
|
||||
receipt = {
|
||||
"started_at": started,
|
||||
"finished_at": utc_now(),
|
||||
"runtime_seconds": runtime,
|
||||
"device": str(device),
|
||||
"torch_version": torch.__version__,
|
||||
"base_model": STUDENT_MODEL,
|
||||
"base_model_revision": STUDENT_REVISION,
|
||||
"training_rows": len(rows),
|
||||
"epochs": epochs,
|
||||
"batch_size": batch_size,
|
||||
"optimizer": "AdamW",
|
||||
"learning_rate": learning_rate,
|
||||
"lora": {"r": 16, "alpha": 32, "targets": ["q_proj", "v_proj"]},
|
||||
"trainable_parameters": sum(p.numel() for p in params),
|
||||
"total_parameters_with_adapter": sum(p.numel() for p in model.parameters()),
|
||||
"optimizer_steps": len(losses),
|
||||
"first_loss": losses[0],
|
||||
"final_loss": losses[-1],
|
||||
"mean_loss": statistics.mean(losses),
|
||||
"losses": losses,
|
||||
}
|
||||
write_json(run_dir / "training_receipt.json", receipt)
|
||||
del model, base
|
||||
if device.type == "mps":
|
||||
torch.mps.empty_cache()
|
||||
elif device.type == "cuda":
|
||||
torch.cuda.empty_cache()
|
||||
return receipt
|
||||
|
||||
|
||||
def _sync(device: Any) -> None:
|
||||
import torch
|
||||
|
||||
if device.type == "mps":
|
||||
torch.mps.synchronize()
|
||||
elif device.type == "cuda":
|
||||
torch.cuda.synchronize(device)
|
||||
|
||||
|
||||
def summarize_retained_teacher_receipts(run_dir: Path) -> dict[str, Any]:
|
||||
"""Summarize durable teacher receipts, including interrupted campaigns."""
|
||||
expected = read_jsonl(run_dir / "benchmark_train_gold.jsonl") + read_jsonl(
|
||||
run_dir / "benchmark_test_gold.jsonl"
|
||||
)
|
||||
expected_by_split = {
|
||||
split: [row for row in expected if row["split"] == split]
|
||||
for split in ("train", "test")
|
||||
}
|
||||
receipts = read_jsonl(run_dir / "teacher_receipts.jsonl")
|
||||
|
||||
def split_summary(split: str) -> dict[str, Any]:
|
||||
expected_rows = expected_by_split[split]
|
||||
rs = [r for r in receipts if r.get("split") == split]
|
||||
valid = [r for r in rs if not r.get("error") and r.get("prediction")]
|
||||
costs = [r.get("calculated_cost") or {} for r in valid]
|
||||
correct = sum(r.get("prediction") == r.get("gold_label") for r in valid)
|
||||
return {
|
||||
"rows": len(expected_rows),
|
||||
"retained_receipts": len(rs),
|
||||
"coverage": len(valid) / len(expected_rows) if expected_rows else 0.0,
|
||||
"valid_receipts": len(valid),
|
||||
"unique_response_ids": len({r.get("response_id") for r in valid}),
|
||||
"correct": correct,
|
||||
# Missing calls do not silently disappear from the campaign score.
|
||||
"gold_accuracy": correct / len(expected_rows) if expected_rows else 0.0,
|
||||
"observed_gold_accuracy": correct / len(valid) if valid else 0.0,
|
||||
"prompt_tokens": sum((r.get("usage") or {}).get("prompt_tokens", 0) for r in valid),
|
||||
"completion_tokens": sum((r.get("usage") or {}).get("completion_tokens", 0) for r in valid),
|
||||
"latency_seconds_total": sum(r.get("latency_seconds", 0) for r in valid),
|
||||
"latency_seconds_mean": statistics.mean(r.get("latency_seconds", 0) for r in valid) if valid else None,
|
||||
"provider_cost_cny": sum(c.get("cny", 0) for c in costs),
|
||||
"provider_cost_usd": sum(c.get("usd", 0) for c in costs),
|
||||
}
|
||||
|
||||
summary = {
|
||||
"teacher": {"provider": "moonshot", "model": TEACHER_MODEL, "pricing": PRICING},
|
||||
"train": split_summary("train"),
|
||||
"test": split_summary("test"),
|
||||
"campaign_complete": len(receipts) == len(expected)
|
||||
and all(not r.get("error") and r.get("response_id") for r in receipts),
|
||||
}
|
||||
write_json(run_dir / "teacher_summary.json", summary)
|
||||
return summary
|
||||
|
||||
|
||||
def evaluate_local_arm(run_dir: Path, arm: str) -> dict[str, Any]:
|
||||
import torch
|
||||
from peft import PeftModel
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
|
||||
device = _local_device(torch)
|
||||
tokenizer = AutoTokenizer.from_pretrained(STUDENT_MODEL, revision=STUDENT_REVISION)
|
||||
if tokenizer.pad_token_id is None:
|
||||
tokenizer.pad_token = tokenizer.eos_token
|
||||
base = AutoModelForCausalLM.from_pretrained(
|
||||
STUDENT_MODEL,
|
||||
revision=STUDENT_REVISION,
|
||||
dtype=torch.bfloat16 if device.type == "cuda" else torch.float32,
|
||||
).to(device)
|
||||
model = base if arm == "baseline" else PeftModel.from_pretrained(base, run_dir / "student_adapter").to(device)
|
||||
model.eval()
|
||||
rows = read_jsonl(run_dir / "benchmark_test_gold.jsonl")
|
||||
|
||||
# Warmup is measured separately and excluded from reported case latency.
|
||||
warm = tokenizer.apply_chat_template(
|
||||
[{"role": "user", "content": "Hello world."}], tokenize=True, add_generation_prompt=True, return_tensors="pt"
|
||||
)
|
||||
if isinstance(warm, dict) or hasattr(warm, "keys"):
|
||||
warm = warm["input_ids"]
|
||||
warm = warm.to(device)
|
||||
with torch.no_grad():
|
||||
_ = model.generate(warm, max_new_tokens=4, do_sample=False, pad_token_id=tokenizer.eos_token_id)
|
||||
_sync(device)
|
||||
|
||||
results: list[dict[str, Any]] = []
|
||||
for row in rows:
|
||||
ids = tokenizer.apply_chat_template(
|
||||
[{"role": "user", "content": row["text"]}],
|
||||
tokenize=True,
|
||||
add_generation_prompt=True,
|
||||
return_tensors="pt",
|
||||
)
|
||||
if isinstance(ids, dict) or hasattr(ids, "keys"):
|
||||
ids = ids["input_ids"]
|
||||
ids = ids.to(device)
|
||||
_sync(device)
|
||||
t0 = time.perf_counter()
|
||||
with torch.no_grad():
|
||||
out = model.generate(
|
||||
ids,
|
||||
max_new_tokens=8,
|
||||
do_sample=False,
|
||||
pad_token_id=tokenizer.eos_token_id,
|
||||
)
|
||||
_sync(device)
|
||||
latency = time.perf_counter() - t0
|
||||
generated = out[0, ids.shape[1] :]
|
||||
text = tokenizer.decode(generated, skip_special_tokens=True).strip()
|
||||
pred = parse_label(text) or (text.lower() if text.lower() in VALID_LABELS else None)
|
||||
results.append(
|
||||
{
|
||||
"id": row["id"],
|
||||
"gold_label": row["gold_label"],
|
||||
"source_label": row["source_label"],
|
||||
"prediction": pred,
|
||||
"response": text,
|
||||
"correct": pred == row["gold_label"],
|
||||
"input_tokens": int(ids.shape[1]),
|
||||
"output_tokens": int(generated.shape[0]),
|
||||
"latency_seconds": latency,
|
||||
}
|
||||
)
|
||||
output_path = run_dir / f"student_{arm}_heldout.jsonl"
|
||||
write_jsonl(output_path, results)
|
||||
summary = {
|
||||
"arm": arm,
|
||||
"rows": len(results),
|
||||
"correct": sum(r["correct"] for r in results),
|
||||
"accuracy": sum(r["correct"] for r in results) / len(results),
|
||||
"parse_rate": sum(r["prediction"] is not None for r in results) / len(results),
|
||||
"input_tokens": sum(r["input_tokens"] for r in results),
|
||||
"output_tokens": sum(r["output_tokens"] for r in results),
|
||||
"latency_seconds_total": sum(r["latency_seconds"] for r in results),
|
||||
"latency_seconds_mean": statistics.mean(r["latency_seconds"] for r in results),
|
||||
"latency_seconds_median": statistics.median(r["latency_seconds"] for r in results),
|
||||
"provider_charge_usd": 0.0,
|
||||
"provider_charge_note": "Local inference made no provider API calls; electricity/hardware amortization is excluded.",
|
||||
}
|
||||
write_json(run_dir / f"student_{arm}_summary.json", summary)
|
||||
del model, base
|
||||
if device.type == "mps":
|
||||
torch.mps.empty_cache()
|
||||
elif device.type == "cuda":
|
||||
torch.cuda.empty_cache()
|
||||
return summary
|
||||
|
||||
|
||||
def package_evidence(run_dir: Path, command: str) -> dict[str, Any]:
|
||||
teacher = summarize_retained_teacher_receipts(run_dir)
|
||||
baseline = json.loads((run_dir / "student_baseline_summary.json").read_text(encoding="utf-8"))
|
||||
trained = json.loads((run_dir / "student_trained_summary.json").read_text(encoding="utf-8"))
|
||||
train_rows = read_jsonl(run_dir / "benchmark_train_gold.jsonl")
|
||||
test_rows = read_jsonl(run_dir / "benchmark_test_gold.jsonl")
|
||||
train_ids = {r["text"] for r in train_rows}
|
||||
test_ids = {r["text"] for r in test_rows}
|
||||
teacher_test = [r for r in read_jsonl(run_dir / "teacher_receipts.jsonl") if r["split"] == "test"]
|
||||
trained_rows = {r["id"]: r for r in read_jsonl(run_dir / "student_trained_heldout.jsonl")}
|
||||
agreement = sum(
|
||||
trained_rows[r["id"]]["prediction"] == r.get("prediction")
|
||||
for r in teacher_test
|
||||
if r["id"] in trained_rows and r.get("prediction")
|
||||
) / len(teacher_test)
|
||||
comparison = {
|
||||
"heldout_rows": len(test_rows),
|
||||
"train_test_exact_text_overlap": len(train_ids & test_ids),
|
||||
"teacher_gold_accuracy": teacher["test"]["gold_accuracy"],
|
||||
"baseline_gold_accuracy": baseline["accuracy"],
|
||||
"trained_gold_accuracy": trained["accuracy"],
|
||||
"trained_absolute_uplift": trained["accuracy"] - baseline["accuracy"],
|
||||
"trained_teacher_agreement": agreement,
|
||||
"teacher_mean_latency_seconds": teacher["test"]["latency_seconds_mean"],
|
||||
"trained_mean_latency_seconds": trained["latency_seconds_mean"],
|
||||
"latency_speedup": teacher["test"]["latency_seconds_mean"] / trained["latency_seconds_mean"],
|
||||
"teacher_input_tokens": teacher["test"]["prompt_tokens"],
|
||||
"trained_input_tokens": trained["input_tokens"],
|
||||
"input_token_reduction": 1 - trained["input_tokens"] / teacher["test"]["prompt_tokens"],
|
||||
"teacher_provider_cost_usd": teacher["test"]["provider_cost_usd"],
|
||||
"trained_provider_charge_usd": trained["provider_charge_usd"],
|
||||
"cost_scope_note": trained["provider_charge_note"],
|
||||
}
|
||||
write_json(run_dir / "comparison.json", comparison)
|
||||
required_files = [
|
||||
p
|
||||
for p in run_dir.rglob("*")
|
||||
if p.is_file() and p.name != "manifest.json"
|
||||
]
|
||||
artifact_hashes = {
|
||||
str(p.relative_to(run_dir)): {"sha256": sha256_file(p), "bytes": p.stat().st_size}
|
||||
for p in sorted(required_files)
|
||||
}
|
||||
receipts = read_jsonl(run_dir / "teacher_receipts.jsonl")
|
||||
gates = {
|
||||
"disjoint_heldout": len(train_ids & test_ids) == 0,
|
||||
"real_teacher_receipts": len(receipts) == len(train_rows) + len(test_rows)
|
||||
and len({r.get("response_id") for r in receipts}) == len(receipts),
|
||||
"real_parameter_training": (run_dir / "student_adapter" / "adapter_model.safetensors").exists(),
|
||||
"before_after_quality": trained["accuracy"] > baseline["accuracy"],
|
||||
"quality_near_teacher": trained["accuracy"] >= teacher["test"]["gold_accuracy"] - 0.15,
|
||||
"measured_latency": teacher["test"]["latency_seconds_mean"] > 0 and trained["latency_seconds_mean"] > 0,
|
||||
"measured_tokens": teacher["test"]["prompt_tokens"] > 0 and trained["input_tokens"] > 0,
|
||||
"dollar_accounting": teacher["test"]["provider_cost_usd"] > 0,
|
||||
}
|
||||
manifest = {
|
||||
"schema_version": 1,
|
||||
"experiment_id": EXPERIMENT_ID,
|
||||
"status": "complete" if all(gates.values()) else "incomplete",
|
||||
"created_at": utc_now(),
|
||||
"command": command,
|
||||
"host": {"platform": sys.platform, "python": sys.version, "credential_env": TEACHER_KEY_ENV},
|
||||
"student": {"model": STUDENT_MODEL, "revision": STUDENT_REVISION},
|
||||
"teacher": {"model": TEACHER_MODEL, "base_url": TEACHER_BASE_URL, "pricing": PRICING},
|
||||
"gates": gates,
|
||||
"comparison": comparison,
|
||||
"artifacts": artifact_hashes,
|
||||
"credential_values_retained": False,
|
||||
}
|
||||
write_json(run_dir / "manifest.json", manifest)
|
||||
return manifest
|
||||
|
||||
|
||||
def main() -> None:
|
||||
parser = argparse.ArgumentParser(description="Run canonical Experiment 8-8")
|
||||
parser.add_argument("--run-dir", default="./validation/exp8-8-real")
|
||||
parser.add_argument("--phase", choices=["all", "prepare", "teacher", "train", "evaluate", "package"], default="all")
|
||||
parser.add_argument("--train-per-language", type=int, default=8)
|
||||
parser.add_argument("--test-per-language", type=int, default=4)
|
||||
parser.add_argument("--seed", type=int, default=78)
|
||||
parser.add_argument("--concurrency", type=int, default=12)
|
||||
parser.add_argument("--max-retries", type=int, default=2)
|
||||
parser.add_argument("--epochs", type=int, default=5)
|
||||
parser.add_argument("--batch-size", type=int, default=4)
|
||||
parser.add_argument("--learning-rate", type=float, default=8e-4)
|
||||
args = parser.parse_args()
|
||||
run_dir = Path(args.run_dir).resolve()
|
||||
run_dir.mkdir(parents=True, exist_ok=True)
|
||||
phases = {args.phase} if args.phase != "all" else {"prepare", "teacher", "train", "evaluate", "package"}
|
||||
if "prepare" in phases:
|
||||
prepare_benchmark(run_dir, args.train_per_language, args.test_per_language, args.seed)
|
||||
if "teacher" in phases:
|
||||
asyncio.run(collect_teacher(run_dir, args.concurrency, args.max_retries))
|
||||
if "train" in phases:
|
||||
train_student(run_dir, args.epochs, args.batch_size, args.learning_rate, args.seed)
|
||||
if "evaluate" in phases:
|
||||
evaluate_local_arm(run_dir, "baseline")
|
||||
evaluate_local_arm(run_dir, "trained")
|
||||
if "package" in phases:
|
||||
command = " ".join([sys.executable, *sys.argv])
|
||||
manifest = package_evidence(run_dir, command)
|
||||
print(json.dumps({"run_dir": str(run_dir), "status": manifest["status"], "comparison": manifest["comparison"]}, indent=2))
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
@@ -0,0 +1,16 @@
|
||||
from compare import compare
|
||||
|
||||
|
||||
def test_compare_empty_texts():
|
||||
result = compare(
|
||||
prompt_template="Classify: {text}",
|
||||
texts=[],
|
||||
teacher_labels={},
|
||||
eval_results=None,
|
||||
count_tokens=lambda s: max(1, len(s) // 4),
|
||||
token_method="test",
|
||||
num_examples=3,
|
||||
)
|
||||
assert result["teacher_input_avg"] == 0.0
|
||||
assert result["student_input_avg"] == 0.0
|
||||
assert result["input_token_reduction_pct"] == 0.0
|
||||
@@ -0,0 +1,37 @@
|
||||
import asyncio
|
||||
|
||||
from create_data import generate_distillation_data
|
||||
|
||||
|
||||
def test_generate_empty_input_file(tmp_path):
|
||||
# 空输入文件（0 个句子）：应干净地直接返回，不加载模型、不抛 IndexError/ZeroDivisionError
|
||||
input_file = tmp_path / "empty.txt"
|
||||
input_file.write_text("", encoding="utf-8")
|
||||
output_file = tmp_path / "out.jsonl"
|
||||
|
||||
result = asyncio.run(
|
||||
generate_distillation_data(
|
||||
input_file=str(input_file),
|
||||
output_file=str(output_file),
|
||||
model_name="stub",
|
||||
)
|
||||
)
|
||||
assert result is None
|
||||
assert not output_file.exists()
|
||||
|
||||
|
||||
def test_generate_blank_lines_only(tmp_path):
|
||||
# 只有空白行的文件同样视为空
|
||||
input_file = tmp_path / "blank.txt"
|
||||
input_file.write_text("\n \n\t\n", encoding="utf-8")
|
||||
output_file = tmp_path / "out.jsonl"
|
||||
|
||||
result = asyncio.run(
|
||||
generate_distillation_data(
|
||||
input_file=str(input_file),
|
||||
output_file=str(output_file),
|
||||
model_name="stub",
|
||||
)
|
||||
)
|
||||
assert result is None
|
||||
assert not output_file.exists()
|
||||
@@ -0,0 +1,25 @@
|
||||
"""Regression: empty test set must not ZeroDivisionError on parse-rate summary."""
|
||||
import sys
|
||||
import types
|
||||
|
||||
|
||||
def _stub_evaluate_deps() -> None:
|
||||
for name in ["torch", "numpy", "transformers", "peft", "tqdm"]:
|
||||
sys.modules.setdefault(name, types.ModuleType(name))
|
||||
sys.modules["transformers"].AutoTokenizer = object
|
||||
sys.modules["transformers"].AutoModelForCausalLM = object
|
||||
sys.modules["peft"].PeftModel = object
|
||||
sys.modules["tqdm"].tqdm = lambda x, **k: x
|
||||
|
||||
|
||||
_stub_evaluate_deps()
|
||||
|
||||
from evaluate import compute_parse_rate # noqa: E402
|
||||
|
||||
|
||||
def test_compute_parse_rate_empty_total_is_zero():
|
||||
assert compute_parse_rate(0, 0) == 0.0
|
||||
|
||||
|
||||
def test_compute_parse_rate_nonzero_total():
|
||||
assert compute_parse_rate(8, 10) == 0.8
|
||||
@@ -0,0 +1,38 @@
|
||||
"""Regression: unparseable model output (pred_label None) must not crash progress prints."""
|
||||
import sys
|
||||
import types
|
||||
|
||||
import pytest
|
||||
|
||||
|
||||
def _stub_evaluate_deps() -> None:
|
||||
for name in ["torch", "numpy", "transformers", "peft", "tqdm"]:
|
||||
sys.modules.setdefault(name, types.ModuleType(name))
|
||||
sys.modules["transformers"].AutoTokenizer = object
|
||||
sys.modules["transformers"].AutoModelForCausalLM = object
|
||||
sys.modules["peft"].PeftModel = object
|
||||
sys.modules["tqdm"].tqdm = lambda x, **k: x
|
||||
|
||||
|
||||
_stub_evaluate_deps()
|
||||
|
||||
from evaluate import format_pred_label, parse_language_label # noqa: E402
|
||||
|
||||
|
||||
def test_parse_language_label_returns_none_for_prose():
|
||||
assert parse_language_label("I believe this is English.") is None
|
||||
|
||||
|
||||
def test_format_pred_label_none_is_displayable():
|
||||
pred_label = parse_language_label("I believe this is English.")
|
||||
assert pred_label is None
|
||||
token = format_pred_label(pred_label)
|
||||
assert token == "??"
|
||||
assert f"Pred: {token:>2s}" == "Pred: ??"
|
||||
with pytest.raises(TypeError):
|
||||
f"{pred_label:>2s}"
|
||||
|
||||
|
||||
def test_format_pred_label_keeps_real_codes():
|
||||
assert format_pred_label("en") == "en"
|
||||
assert f"{format_pred_label('fr'):>2s}" == "fr"
|
||||
@@ -0,0 +1,65 @@
|
||||
import importlib.util
|
||||
import sys
|
||||
from pathlib import Path
|
||||
|
||||
|
||||
MODULE_PATH = Path(__file__).with_name("run_experiment_8_8.py")
|
||||
SPEC = importlib.util.spec_from_file_location("experiment_8_8", MODULE_PATH)
|
||||
exp = importlib.util.module_from_spec(SPEC)
|
||||
assert SPEC and SPEC.loader
|
||||
sys.modules[SPEC.name] = exp
|
||||
SPEC.loader.exec_module(exp)
|
||||
|
||||
|
||||
def test_parse_label_is_strict():
|
||||
assert exp.parse_label("Final Answer: fr") == "fr"
|
||||
assert exp.parse_label("zh") == "zh"
|
||||
assert exp.parse_label("Final Answer: jp") is None
|
||||
assert exp.parse_label("The answer is French") is None
|
||||
|
||||
|
||||
def test_cost_uses_uncached_cached_and_output_rates():
|
||||
usage = {
|
||||
"prompt_tokens": 100,
|
||||
"completion_tokens": 10,
|
||||
"prompt_tokens_details": {"cached_tokens": 40},
|
||||
}
|
||||
cost = exp.usage_cost(usage)
|
||||
expected_cny = (60 * 20 + 40 * 2 + 10 * 100) / 1_000_000
|
||||
assert cost["cny"] == expected_cny
|
||||
assert cost["usd"] == expected_cny * exp.PRICING["usd_per_currency_unit"]
|
||||
|
||||
|
||||
def test_stable_sample_maps_out_of_scope_languages_to_other():
|
||||
rows = [
|
||||
{"labels": "fr", "text": "bonjour"},
|
||||
{"labels": "ja", "text": "こんにちは"},
|
||||
]
|
||||
sampled = exp._stable_sample(rows, n=1, seed=1, split="train")
|
||||
by_source = {row["source_label"]: row for row in sampled}
|
||||
assert by_source["fr"]["gold_label"] == "fr"
|
||||
assert by_source["ja"]["gold_label"] == "ot"
|
||||
|
||||
|
||||
def test_sft_rows_expose_only_raw_text_to_student():
|
||||
row = {
|
||||
"text": "bonjour",
|
||||
"prediction": "fr",
|
||||
"id": "x",
|
||||
"gold_label": "fr",
|
||||
"response_id": "receipt",
|
||||
}
|
||||
sample = {
|
||||
"id": row["id"],
|
||||
"messages": [
|
||||
{"role": "user", "content": row["text"]},
|
||||
{"role": "assistant", "content": row["prediction"]},
|
||||
],
|
||||
}
|
||||
assert exp.LANGUAGE_CLASSIFICATION_PROMPT not in sample["messages"][0]["content"]
|
||||
assert sample["messages"][0]["content"] == "bonjour"
|
||||
|
||||
|
||||
def test_chat_template_ids_accepts_transformers_4_and_5_shapes():
|
||||
assert exp._chat_template_ids([1, 2, 3]) == [1, 2, 3]
|
||||
assert exp._chat_template_ids({"input_ids": [1, 2, 3], "attention_mask": [1, 1, 1]}) == [1, 2, 3]
|
||||
@@ -0,0 +1,490 @@
|
||||
"""
|
||||
Prompt Distillation Training using Hugging Face TRL
|
||||
|
||||
This script trains a student model using the TRL SFTTrainer instead of verl.
|
||||
TRL is more widely used, better documented, and easier to work with.
|
||||
|
||||
Based on the same prompt distillation methodology but using standard HF tools.
|
||||
"""
|
||||
|
||||
import argparse
|
||||
import json
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import torch
|
||||
from datasets import Dataset
|
||||
from transformers import AutoModelForCausalLM, AutoTokenizer
|
||||
from peft import LoraConfig, get_peft_model
|
||||
|
||||
# 注意：trl 的 SFTTrainer / SFTConfig 在 train_model() 内部按需导入，
|
||||
# 这样即便未安装 trl（如离线查看 --help 时）也能正常展示命令行帮助。
|
||||
|
||||
|
||||
def load_jsonl_dataset(file_path: str) -> Dataset:
|
||||
"""
|
||||
Load training data from JSONL file.
|
||||
|
||||
Args:
|
||||
file_path: Path to JSONL file with messages format
|
||||
|
||||
Returns:
|
||||
Dataset: Hugging Face Dataset object
|
||||
"""
|
||||
local_rank = int(os.environ.get("LOCAL_RANK", 0))
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"Loading dataset from: {file_path}")
|
||||
|
||||
data = []
|
||||
with open(file_path, "r", encoding="utf-8") as f:
|
||||
for line in f:
|
||||
if line.strip():
|
||||
data.append(json.loads(line))
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"Loaded {len(data)} training examples")
|
||||
|
||||
# Show sample
|
||||
if data:
|
||||
print(f"\nSample data:")
|
||||
print(f" Messages: {data[0]['messages']}")
|
||||
|
||||
# Convert to HF Dataset
|
||||
dataset = Dataset.from_list(data)
|
||||
|
||||
return dataset
|
||||
|
||||
|
||||
def prepare_model_and_tokenizer(model_name: str, use_lora: bool = True,
|
||||
lora_rank: int = 32, lora_alpha: int = 16):
|
||||
"""
|
||||
Load model and tokenizer, optionally with LoRA.
|
||||
|
||||
Args:
|
||||
model_name: Model name or path
|
||||
use_lora: Whether to use LoRA for efficient training
|
||||
lora_rank: LoRA rank
|
||||
lora_alpha: LoRA alpha parameter
|
||||
|
||||
Returns:
|
||||
tuple: (model, tokenizer, peft_config or None)
|
||||
"""
|
||||
local_rank = int(os.environ.get("LOCAL_RANK", 0))
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"\n{'='*80}")
|
||||
print(f"Loading Model and Tokenizer")
|
||||
print(f"{'='*80}")
|
||||
print(f"Model: {model_name}")
|
||||
print(f"LoRA: {'Enabled' if use_lora else 'Disabled'}")
|
||||
|
||||
# Load tokenizer
|
||||
tokenizer = AutoTokenizer.from_pretrained(model_name, trust_remote_code=True)
|
||||
|
||||
# Set pad token if not exists
|
||||
if tokenizer.pad_token is None:
|
||||
tokenizer.pad_token = tokenizer.eos_token
|
||||
tokenizer.pad_token_id = tokenizer.eos_token_id
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"Tokenizer loaded: vocab_size={len(tokenizer)}")
|
||||
|
||||
# Load model
|
||||
# Note: Don't use device_map='auto' in distributed training - let DDP/FSDP handle device placement
|
||||
model_kwargs = {
|
||||
"torch_dtype": torch.bfloat16,
|
||||
"trust_remote_code": True,
|
||||
"use_cache": False, # Disable for training
|
||||
}
|
||||
|
||||
# Only use device_map for single GPU (non-distributed)
|
||||
if local_rank == -1 or int(os.environ.get("WORLD_SIZE", "1")) == 1:
|
||||
model_kwargs["device_map"] = "auto"
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"Loading model (this may take a few minutes)...")
|
||||
model = AutoModelForCausalLM.from_pretrained(model_name, **model_kwargs)
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"Model loaded successfully!")
|
||||
print(f" Parameters: {model.num_parameters() / 1e9:.2f}B")
|
||||
|
||||
# Configure LoRA if enabled
|
||||
peft_config = None
|
||||
if use_lora:
|
||||
if local_rank == 0:
|
||||
print(f"\nConfiguring LoRA:")
|
||||
print(f" Rank: {lora_rank}")
|
||||
print(f" Alpha: {lora_alpha}")
|
||||
|
||||
peft_config = LoraConfig(
|
||||
r=lora_rank,
|
||||
lora_alpha=lora_alpha,
|
||||
target_modules="all-linear",
|
||||
lora_dropout=0.0,
|
||||
bias="none",
|
||||
task_type="CAUSAL_LM",
|
||||
)
|
||||
|
||||
model = get_peft_model(model, peft_config)
|
||||
|
||||
# Print trainable parameters
|
||||
if local_rank == 0:
|
||||
trainable_params = sum(p.numel() for p in model.parameters() if p.requires_grad)
|
||||
total_params = sum(p.numel() for p in model.parameters())
|
||||
trainable_percent = 100 * trainable_params / total_params
|
||||
|
||||
print(f"\nTrainable Parameters:")
|
||||
print(f" Trainable: {trainable_params:,} ({trainable_percent:.2f}%)")
|
||||
print(f" Total: {total_params:,}")
|
||||
|
||||
return model, tokenizer, peft_config
|
||||
|
||||
|
||||
def train_model(
|
||||
model,
|
||||
tokenizer,
|
||||
train_dataset,
|
||||
output_dir: str,
|
||||
num_train_epochs: int = 1,
|
||||
per_device_train_batch_size: int = 4,
|
||||
gradient_accumulation_steps: int = 4,
|
||||
learning_rate: float = 2e-4,
|
||||
max_length: int = 2048,
|
||||
warmup_ratio: float = 0.03,
|
||||
logging_steps: int = 1,
|
||||
save_strategy: str = "epoch",
|
||||
lr_scheduler_type: str = "cosine_with_min_lr",
|
||||
report_to: str = "wandb",
|
||||
run_name: str = None,
|
||||
):
|
||||
"""
|
||||
Train the model using TRL SFTTrainer.
|
||||
|
||||
Hyperparameters are based on the OpenAI Cookbook gpt-oss-20b example,
|
||||
which provides good defaults for efficient fine-tuning.
|
||||
|
||||
Args:
|
||||
model: The model to train
|
||||
tokenizer: The tokenizer
|
||||
train_dataset: Training dataset
|
||||
output_dir: Output directory for checkpoints
|
||||
num_train_epochs: Number of training epochs (default: 1, matching OpenAI)
|
||||
per_device_train_batch_size: Batch size per device (default: 4, matching OpenAI)
|
||||
gradient_accumulation_steps: Gradient accumulation steps (default: 4, matching OpenAI)
|
||||
learning_rate: Learning rate (default: 2e-4, matching OpenAI)
|
||||
max_length: Maximum sequence length (default: 2048, matching OpenAI)
|
||||
warmup_ratio: Warmup ratio (default: 0.03, matching OpenAI)
|
||||
logging_steps: Steps between logging (default: 1, matching OpenAI)
|
||||
save_strategy: When to save checkpoints
|
||||
lr_scheduler_type: Learning rate scheduler type (default: cosine_with_min_lr, matching OpenAI)
|
||||
report_to: Where to report metrics (default: wandb)
|
||||
run_name: Custom run name for logging (default: auto-generated)
|
||||
|
||||
Returns:
|
||||
SFTTrainer: The trained trainer object
|
||||
"""
|
||||
from trl import SFTTrainer, SFTConfig
|
||||
|
||||
local_rank = int(os.environ.get("LOCAL_RANK", 0))
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"\n{'='*80}")
|
||||
print(f"Training Configuration")
|
||||
print(f"{'='*80}")
|
||||
|
||||
# Calculate effective batch size
|
||||
world_size = torch.cuda.device_count() if torch.cuda.is_available() else 1
|
||||
effective_batch_size = per_device_train_batch_size * gradient_accumulation_steps * world_size
|
||||
|
||||
# Detect distributed mode
|
||||
distributed_mode = "Single GPU"
|
||||
if int(os.environ.get("WORLD_SIZE", "1")) > 1:
|
||||
if os.environ.get("ACCELERATE_USE_FSDP", "false").lower() == "true":
|
||||
distributed_mode = "FSDP (Fully Sharded Data Parallel)"
|
||||
else:
|
||||
distributed_mode = "DDP (Distributed Data Parallel)"
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"Training Parameters:")
|
||||
print(f" Output directory: {output_dir}")
|
||||
print(f" Distributed mode: {distributed_mode}")
|
||||
print(f" Epochs: {num_train_epochs}")
|
||||
print(f" Per-device batch size: {per_device_train_batch_size}")
|
||||
print(f" Gradient accumulation steps: {gradient_accumulation_steps}")
|
||||
print(f" Number of GPUs: {world_size}")
|
||||
print(f" Effective batch size: {effective_batch_size}")
|
||||
print(f" Learning rate: {learning_rate}")
|
||||
print(f" LR scheduler: {lr_scheduler_type}")
|
||||
print(f" Warmup ratio: {warmup_ratio}")
|
||||
print(f" Max length: {max_length}")
|
||||
print(f" Logging steps: {logging_steps}")
|
||||
print(f" Save strategy: {save_strategy}")
|
||||
|
||||
# Show memory advantage for FSDP
|
||||
if "FSDP" in distributed_mode:
|
||||
print(f"\n 💡 FSDP Mode: Each GPU holds ~{100/world_size:.1f}% of the model")
|
||||
|
||||
# Training configuration (matching OpenAI Cookbook gpt-oss-20b example)
|
||||
training_args = SFTConfig(
|
||||
output_dir=output_dir,
|
||||
num_train_epochs=num_train_epochs,
|
||||
per_device_train_batch_size=per_device_train_batch_size,
|
||||
gradient_accumulation_steps=gradient_accumulation_steps,
|
||||
learning_rate=learning_rate,
|
||||
max_length=max_length, # Note: Use max_length, not max_seq_length
|
||||
warmup_ratio=warmup_ratio,
|
||||
lr_scheduler_type=lr_scheduler_type,
|
||||
lr_scheduler_kwargs={"min_lr_rate": 0.1} if lr_scheduler_type == "cosine_with_min_lr" else {},
|
||||
logging_steps=logging_steps,
|
||||
save_strategy=save_strategy,
|
||||
save_total_limit=2, # Keep only last 2 checkpoints
|
||||
gradient_checkpointing=True, # Save memory (matching OpenAI)
|
||||
bf16=torch.cuda.is_available(), # Use bfloat16 if available
|
||||
logging_first_step=True,
|
||||
report_to=report_to, # wandb, tensorboard, or none
|
||||
run_name=run_name or f"prompt-distillation-{num_train_epochs}epoch",
|
||||
remove_unused_columns=False,
|
||||
dataset_text_field="", # We'll use formatting function
|
||||
dataset_kwargs={
|
||||
"skip_prepare_dataset": False,
|
||||
},
|
||||
)
|
||||
|
||||
if local_rank == 0:
|
||||
if report_to == "wandb":
|
||||
print(f"\n 📊 Logging to Weights & Biases (wandb)")
|
||||
print(f" Run name: {run_name or f'prompt-distillation-{num_train_epochs}epoch'}")
|
||||
print(f" View at: https://wandb.ai")
|
||||
elif report_to == "tensorboard":
|
||||
print(f"\n 📊 Logging to TensorBoard")
|
||||
print(f" View with: tensorboard --logdir {output_dir}")
|
||||
else:
|
||||
print(f"\n 📊 Logging disabled (report_to=none)")
|
||||
|
||||
# Initialize trainer
|
||||
if local_rank == 0:
|
||||
print(f"\n{'='*80}")
|
||||
print(f"Initializing SFTTrainer")
|
||||
print(f"{'='*80}")
|
||||
|
||||
trainer = SFTTrainer(
|
||||
model=model,
|
||||
args=training_args,
|
||||
train_dataset=train_dataset,
|
||||
processing_class=tokenizer,
|
||||
formatting_func=lambda x: tokenizer.apply_chat_template(
|
||||
x["messages"],
|
||||
tokenize=False,
|
||||
add_generation_prompt=False,
|
||||
),
|
||||
)
|
||||
|
||||
# Start training
|
||||
if local_rank == 0:
|
||||
print(f"\n{'='*80}")
|
||||
print(f"Starting Training")
|
||||
print(f"{'='*80}")
|
||||
|
||||
trainer.train()
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"\n{'='*80}")
|
||||
print(f"Training Complete!")
|
||||
print(f"{'='*80}")
|
||||
|
||||
return trainer
|
||||
|
||||
|
||||
def main():
|
||||
parser = argparse.ArgumentParser(
|
||||
description="用 Hugging Face TRL 训练 Prompt 蒸馏学生模型（无提示直接作答）",
|
||||
formatter_class=argparse.ArgumentDefaultsHelpFormatter,
|
||||
)
|
||||
|
||||
# Data arguments
|
||||
parser.add_argument(
|
||||
"--train_file",
|
||||
type=str,
|
||||
default="./data/prompt_distillation_lang.jsonl",
|
||||
help="训练数据路径（JSONL 格式）",
|
||||
)
|
||||
|
||||
# Model arguments
|
||||
parser.add_argument(
|
||||
"--model_name",
|
||||
type=str,
|
||||
default="Qwen/Qwen3-30B-A3B-Instruct-2507",
|
||||
help="学生基座模型名称或路径（用于蒸馏的非思考型模型）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--output_dir",
|
||||
type=str,
|
||||
default="./models/prompt_distillation_trl",
|
||||
help="模型 checkpoint 的输出目录",
|
||||
)
|
||||
|
||||
# LoRA arguments
|
||||
parser.add_argument(
|
||||
"--use_lora",
|
||||
action="store_true",
|
||||
default=True,
|
||||
help="使用 LoRA 做参数高效微调",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--lora_rank",
|
||||
type=int,
|
||||
default=32,
|
||||
help="LoRA rank（默认 32，与 tinker 一致）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--lora_alpha",
|
||||
type=int,
|
||||
default=16,
|
||||
help="LoRA alpha 参数（默认 16）",
|
||||
)
|
||||
|
||||
# Training arguments
|
||||
parser.add_argument(
|
||||
"--num_train_epochs",
|
||||
type=int,
|
||||
default=1,
|
||||
help="训练轮数（默认 1，与 OpenAI 教程一致）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--per_device_train_batch_size",
|
||||
type=int,
|
||||
default=4,
|
||||
help="每张 GPU 的批次大小（默认 4，与 OpenAI 教程一致）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--gradient_accumulation_steps",
|
||||
type=int,
|
||||
default=4,
|
||||
help="梯度累积步数（默认 4，与 OpenAI 教程一致）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--learning_rate",
|
||||
type=float,
|
||||
default=2e-4,
|
||||
help="学习率（默认 2e-4，与 OpenAI 教程一致）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--max_length",
|
||||
type=int,
|
||||
default=2048,
|
||||
help="最大序列长度（默认 2048，与 OpenAI 教程一致）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--warmup_ratio",
|
||||
type=float,
|
||||
default=0.03,
|
||||
help="warmup 比例（默认 0.03，与 OpenAI 教程一致）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--lr_scheduler_type",
|
||||
type=str,
|
||||
default="cosine_with_min_lr",
|
||||
help="学习率调度器类型（默认 cosine_with_min_lr，与 OpenAI 教程一致）",
|
||||
)
|
||||
|
||||
# Logging arguments
|
||||
parser.add_argument(
|
||||
"--report_to",
|
||||
type=str,
|
||||
default="wandb",
|
||||
choices=["wandb", "tensorboard", "none"],
|
||||
help="训练指标上报目标（默认 wandb）",
|
||||
)
|
||||
parser.add_argument(
|
||||
"--run_name",
|
||||
type=str,
|
||||
default=None,
|
||||
help="wandb 运行名称（未提供则自动生成）",
|
||||
)
|
||||
|
||||
args = parser.parse_args()
|
||||
|
||||
# Print configuration
|
||||
print(f"{'='*80}")
|
||||
print(f"Prompt Distillation Training with TRL")
|
||||
print(f"{'='*80}")
|
||||
print(f"Configuration:")
|
||||
print(f" Train file: {args.train_file}")
|
||||
print(f" Model: {args.model_name}")
|
||||
print(f" Output dir: {args.output_dir}")
|
||||
print(f" LoRA: {args.use_lora}")
|
||||
if args.use_lora:
|
||||
print(f" - Rank: {args.lora_rank}")
|
||||
print(f" - Alpha: {args.lora_alpha}")
|
||||
print(f" Epochs: {args.num_train_epochs}")
|
||||
print(f" Batch size: {args.per_device_train_batch_size}")
|
||||
print(f" Gradient accumulation: {args.gradient_accumulation_steps}")
|
||||
print(f" Learning rate: {args.learning_rate}")
|
||||
print(f" Max length: {args.max_length}")
|
||||
print(f" LR scheduler: {args.lr_scheduler_type}")
|
||||
print(f" Warmup ratio: {args.warmup_ratio}")
|
||||
print(f"{'='*80}\n")
|
||||
|
||||
# Check if training file exists
|
||||
if not os.path.exists(args.train_file):
|
||||
raise FileNotFoundError(f"Training file not found: {args.train_file}")
|
||||
|
||||
# Create output directory
|
||||
os.makedirs(args.output_dir, exist_ok=True)
|
||||
|
||||
# Load dataset
|
||||
train_dataset = load_jsonl_dataset(args.train_file)
|
||||
|
||||
# Load model and tokenizer
|
||||
model, tokenizer, peft_config = prepare_model_and_tokenizer(
|
||||
args.model_name,
|
||||
use_lora=args.use_lora,
|
||||
lora_rank=args.lora_rank,
|
||||
lora_alpha=args.lora_alpha,
|
||||
)
|
||||
|
||||
# Train model
|
||||
trainer = train_model(
|
||||
model=model,
|
||||
tokenizer=tokenizer,
|
||||
train_dataset=train_dataset,
|
||||
output_dir=args.output_dir,
|
||||
num_train_epochs=args.num_train_epochs,
|
||||
per_device_train_batch_size=args.per_device_train_batch_size,
|
||||
gradient_accumulation_steps=args.gradient_accumulation_steps,
|
||||
learning_rate=args.learning_rate,
|
||||
max_length=args.max_length,
|
||||
warmup_ratio=args.warmup_ratio,
|
||||
lr_scheduler_type=args.lr_scheduler_type,
|
||||
report_to=args.report_to,
|
||||
run_name=args.run_name,
|
||||
)
|
||||
|
||||
# Save final model (only main process)
|
||||
local_rank = int(os.environ.get("LOCAL_RANK", 0))
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"\nSaving final model to: {args.output_dir}")
|
||||
trainer.save_model(args.output_dir)
|
||||
tokenizer.save_pretrained(args.output_dir)
|
||||
|
||||
if local_rank == 0:
|
||||
print(f"\n{'='*80}")
|
||||
print(f"Training Complete!")
|
||||
print(f"{'='*80}")
|
||||
print(f"Model saved to: {args.output_dir}")
|
||||
print(f"\nTo use the model:")
|
||||
print(f" from transformers import AutoModelForCausalLM, AutoTokenizer")
|
||||
print(f" from peft import PeftModel")
|
||||
print(f" ")
|
||||
print(f" tokenizer = AutoTokenizer.from_pretrained('{args.output_dir}')")
|
||||
print(f" model = AutoModelForCausalLM.from_pretrained('{args.model_name}')")
|
||||
print(f" model = PeftModel.from_pretrained(model, '{args.output_dir}')")
|
||||
print(f"{'='*80}\n")
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
|
||||
Executable
+55
@@ -0,0 +1,55 @@
|
||||
#!/bin/bash
|
||||
# Training script using Hugging Face TRL (more common and easier to use than verl)
|
||||
#
|
||||
# This script uses the widely-adopted TRL SFTTrainer instead of verl.
|
||||
# Benefits:
|
||||
# - Works directly with JSONL (no parquet conversion needed)
|
||||
# - Better documentation and community support
|
||||
# - Simpler setup and fewer dependencies
|
||||
# - Standard Hugging Face workflow
|
||||
|
||||
set -x
|
||||
|
||||
# Default configuration
|
||||
MODEL_NAME=${1:-"Qwen/Qwen3-30B-A3B-Instruct-2507"}
|
||||
OUTPUT_DIR=${2:-"./models/prompt_distillation_trl"}
|
||||
TRAIN_FILE=${3:-"./data/prompt_distillation_lang.jsonl"}
|
||||
|
||||
echo "============================================"
|
||||
echo "Prompt Distillation Training with TRL"
|
||||
echo "============================================"
|
||||
echo "Model: $MODEL_NAME"
|
||||
echo "Output: $OUTPUT_DIR"
|
||||
echo "Train file: $TRAIN_FILE"
|
||||
echo "============================================"
|
||||
echo ""
|
||||
|
||||
# Check if training file exists
|
||||
if [ ! -f "$TRAIN_FILE" ]; then
|
||||
echo "❌ Training file not found: $TRAIN_FILE"
|
||||
echo "Please run data generation first:"
|
||||
echo " python create_data.py"
|
||||
exit 1
|
||||
fi
|
||||
|
||||
# Run training with OpenAI-style hyperparameters
|
||||
python train_sft_trl.py \
|
||||
--model_name "$MODEL_NAME" \
|
||||
--output_dir "$OUTPUT_DIR" \
|
||||
--train_file "$TRAIN_FILE" \
|
||||
--use_lora \
|
||||
--lora_rank 32 \
|
||||
--lora_alpha 16 \
|
||||
--num_train_epochs 1 \
|
||||
--per_device_train_batch_size 4 \
|
||||
--gradient_accumulation_steps 4 \
|
||||
--learning_rate 2e-4 \
|
||||
--max_length 2048 \
|
||||
--warmup_ratio 0.03 \
|
||||
--lr_scheduler_type cosine_with_min_lr
|
||||
|
||||
echo ""
|
||||
echo "============================================"
|
||||
echo "✅ Training Complete!"
|
||||
echo "============================================"
|
||||
echo "Model saved to: $OUTPUT_DIR"
|
||||
+80
@@ -0,0 +1,80 @@
|
||||
{"id": "test-ar-000", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "إن الاستمرارية ، بعد كل شيء ، هي فضيلة ، أو هكذا يقول أولئك الذين لم يفرضوها عليهم."}
|
||||
{"id": "test-ar-001", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "حاولت مكاتب السياحة إعادة تسمية منطقة L'Estrie ، ولكن حتى أكثر الكيوبيين المتشددين تشدد على العرف ، إذا كان تقريبًا ، ترجمة للمقاطعات."}
|
||||
{"id": "test-ar-002", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "حاولت مكاتب السياحة إعادة تسمية منطقة L'Estrie ، ولكن حتى أكثر الكيوبيين المتشددين تشدد على العرف ، إذا كان تقريبًا ، ترجمة للمقاطعات."}
|
||||
{"id": "test-ar-003", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "هذا هو رئيس الرقيب كليم فرانسيس ، متقاعد من القوات الجوية الأمريكية."}
|
||||
{"id": "test-bg-000", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "Освен че проучи внимателно различни официални документи, чешкото правителство разгледа също така снимки от охранителни камери, направени пред посолството на Ирак."}
|
||||
{"id": "test-bg-001", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "Общата биология всъщност е на една ръка разстояние."}
|
||||
{"id": "test-bg-002", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "Тънкостите на галското подчинително наклонение изобщо не го притесняват и той дори не се безпокои, когато се упражнява."}
|
||||
{"id": "test-bg-003", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "80% от участниците ще покажат повишени умения за разрешаване на конфликти."}
|
||||
{"id": "test-de-000", "split": "test", "source_label": "de", "gold_label": "de", "text": "Die Abdeckung (dünnes Plastik)war leider gebrochen! Habe Ersatz angefordert, hat auch super geklappt, nur das die Lampe wieder den selben Schaden hat... wenn mich nicht alles täuscht wurde mir die gleiche einfach nochmal geschickt!!!!😈 Hoffentlich passts beim dritten mal!!!😡 Ansonsten gäbe es bei dem Preis eh nichts auszusetzen! Aber mehr Sterne hat sich der Service nicht verdient..."}
|
||||
{"id": "test-de-001", "split": "test", "source_label": "de", "gold_label": "de", "text": "Besser etwas größer bestellen !"}
|
||||
{"id": "test-de-002", "split": "test", "source_label": "de", "gold_label": "de", "text": "Lichter sind gut. Farben sind OK. Aber es hängt und es Klackert.--.Schade.--!!"}
|
||||
{"id": "test-de-003", "split": "test", "source_label": "de", "gold_label": "de", "text": "Dicht und Robust, alles was er soll."}
|
||||
{"id": "test-el-000", "split": "test", "source_label": "el", "gold_label": "el", "text": "Ήταν η μοναδική απώλεια στην Κρίση της Κούβας και, ο Kaiser, πήρε τις φωτογραφίες και πέταξε κατευθείαν στο Αεροδρόμιο Andrews της Πολεμικής Αεροπορίας στην Ουάσινγκτον."}
|
||||
{"id": "test-el-001", "split": "test", "source_label": "el", "gold_label": "el", "text": "Όμως, καθώς μεγάλωνε, ε, δεν αναγνώρισε ποτέ ότι έκανε λάθος, αλλά άλλαξε τη συμπεριφορά της."}
|
||||
{"id": "test-el-002", "split": "test", "source_label": "el", "gold_label": "el", "text": "Και είναι διαφορετικό, όπως κάτω από κάθε πελάτη, είναι όλα τα αρχεία τους."}
|
||||
{"id": "test-el-003", "split": "test", "source_label": "el", "gold_label": "el", "text": "Και είναι διαφορετικό, όπως κάτω από κάθε πελάτη, είναι όλα τα αρχεία τους."}
|
||||
{"id": "test-en-000", "split": "test", "source_label": "en", "gold_label": "en", "text": "We like the rack. There were some sharp pointed wires that needed to be cut before installing it."}
|
||||
{"id": "test-en-001", "split": "test", "source_label": "en", "gold_label": "en", "text": "Love the gloves , but be,careful if you are allergic to nickel don't buy them they are made with nickel chloride which I am allergic very highly allergic to so that was the only downfall"}
|
||||
{"id": "test-en-002", "split": "test", "source_label": "en", "gold_label": "en", "text": "Ordered this for my son. We waited for 3weeks to arrive. The left side where the whole for the garter is torn so we had to tape it. The black paint was erased after a week. This mask is really thin."}
|
||||
{"id": "test-en-003", "split": "test", "source_label": "en", "gold_label": "en", "text": "Bought a few for craft show tables. I’ve owned them for years & I just can’t get the wrinkles out."}
|
||||
{"id": "test-es-000", "split": "test", "source_label": "es", "gold_label": "es", "text": "La compré para mi mujer y le ha gustado bastante. Es muy grande, pero dormimos en una cama de 1,50m y no hemos tenido problema de espacio. La única pega que encuentra es que es baja para dormir sin almohada, si pone la almohada debajo le resulta demasiado alta. También la usa en el sofá. Los materiales son de calidad y agradables"}
|
||||
{"id": "test-es-001", "split": "test", "source_label": "es", "gold_label": "es", "text": "No se sujeta bien al secador y es un secador universal con boquilla larga. No es satisfactorio para mi agrado"}
|
||||
{"id": "test-es-002", "split": "test", "source_label": "es", "gold_label": "es", "text": "Han tardado el tiempo que decía en la entrega, y todo estaba bien. Pero creo que le falta una guía/soporte, al igual que otros protectores de pantalla tienen para ser más sencillos fe colocar. También creo que no es exactamente igual que el OPPO A3, pero aún así ha quedado bien."}
|
||||
{"id": "test-es-003", "split": "test", "source_label": "es", "gold_label": "es", "text": "Buenas tardes. No me ha llegado el pedido del libro y pone que está entregado. Espero su contestación lo antes posible gracias"}
|
||||
{"id": "test-fr-000", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "Au bout de 1 mois quasiment casser j'ai du re commander une coque. Le miroir est très fragile et le tissu sur le bord aussi"}
|
||||
{"id": "test-fr-001", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "livraison vraiment désastreux par dhl et colissimo voir les photos en dessous"}
|
||||
{"id": "test-fr-002", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "Copie chinoise de l'original. Qualité de finition moyenne, j'ai même un doute sur la sécurité. Le dock ne fonctionne pas avec. pas assez puissant."}
|
||||
{"id": "test-fr-003", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "compliqué ce livre mais après chaque envoi je vous enverrai une satisfaction du moment que le transporteur est bpost plus facile"}
|
||||
{"id": "test-hi-000", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "इसलिए मुझे नहीं पता कि मैंने तमन्ना की थी या नहीं |"}
|
||||
{"id": "test-hi-001", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "इसलिए मुझे नहीं पता कि मैंने तमन्ना की थी या नहीं |"}
|
||||
{"id": "test-hi-002", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "और, ज़ाहिर है, एंड्रोव ग्रोमिकोव ने कोई जवाब नहीं दिया, लेकिन हमारे पास यू 2 की फिल्मों के आधार पर सारी जानकारी थी।"}
|
||||
{"id": "test-hi-003", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "उह, मैं अभी भी एक और केवल नौ दो-दो था जिसने कभी रेगुलेटर पर इंजेक्शन सेट कर दिया था।"}
|
||||
{"id": "test-it-000", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Una persona sta affettando delle cipolle."}
|
||||
{"id": "test-it-001", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Diversi bambini saltano su e giù su un trampolino."}
|
||||
{"id": "test-it-002", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Il tasso di disoccupazione civile è migliorato marginalmente il mese scorso, scendendo al 6,1 per cento, anche se le aziende hanno tagliato le buste paga di 93.000 unità."}
|
||||
{"id": "test-it-003", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Myanmar arresta 44 persone per violenza settaria"}
|
||||
{"id": "test-ja-000", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "SSDへシステムの引っ越しに使いました。 簡単に目的は達成しました。 ソケットのバネが弱いようです。 簡単に外れてしまう ので・・・それなりの工夫が必要でした。 個体差はあるかもしれませんが・・・。"}
|
||||
{"id": "test-ja-001", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "軽いので持ち運びしやすいです。 車の中でも使えることも考慮すると手頃かなと思います。"}
|
||||
{"id": "test-ja-002", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "もう、見るべしとしか言えない。 でも、一部のヒロインが好きな方（僕）には、残念。出てこないんだもん。あんまり。"}
|
||||
{"id": "test-ja-003", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "なんだかちょいちょいバグるのか左に移動します。 が、繋ぎ直すと大丈夫！ でも、ちょっと面倒なので☆１つ減らしときました。"}
|
||||
{"id": "test-nl-000", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "Het vlees, gevogelte, boter, kaas en noten werden een jaar geleden in beslag genomen in een LaGrou Cold Storage warehouse in Chicago."}
|
||||
{"id": "test-nl-001", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "Een zwarte hond die naar de camera kijkt."}
|
||||
{"id": "test-nl-002", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "Een schildpad zwemt in het water."}
|
||||
{"id": "test-nl-003", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "De technologisch geplaatste Nasdaq Composite Index .IXIC daalde met 25,36 punten, of 1,53 procent, tot 1.628,26."}
|
||||
{"id": "test-pl-000", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Opalona dziewczyna w kwiatowym bikini pływa w ciemnoniebieskiej wodzie."}
|
||||
{"id": "test-pl-001", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Jeśli nie jesteś pewien, jak to zrobić, nie rób tego wcale."}
|
||||
{"id": "test-pl-002", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Rada Kolebki Wolności nie jest pierwszą, która złamała stanowisko grupy narodowej."}
|
||||
{"id": "test-pl-003", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Średnia dla przemysłu Dow Jones spadła o 10,89 dnia do poziomu 9 837,94, po awansie o 111,04 w środę."}
|
||||
{"id": "test-pt-000", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "A Kollar-Kotelly agendou outra audiência para Janeiro sobre o cumprimento do acordo antitrust."}
|
||||
{"id": "test-pt-001", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "A jovem mulher está a namoriscar com o jovem rapaz."}
|
||||
{"id": "test-pt-002", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "Um grupo de homens joga futebol na praia."}
|
||||
{"id": "test-pt-003", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "Uma jovem com sujidade na cara e uma bicicleta vermelha de criança está no fundo."}
|
||||
{"id": "test-ru-000", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Моим детям сейчас двадцать один и двадцать четыре года, поэтому, мне не приходится"}
|
||||
{"id": "test-ru-001", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Но я торопился высадить тебя."}
|
||||
{"id": "test-ru-002", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Но я торопился высадить тебя."}
|
||||
{"id": "test-ru-003", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Через каждые сто градусов пятна краски меняют свой цвет, она может быть красной и изменить цвет на синий."}
|
||||
{"id": "test-sw-000", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Baada ya kukataa kwanza ombi la Hazmi ya mkopo, msimamizi alikubali kumruhusu kutumia akaunti ya benki ya msimamizi ili kupokea uhamisho wa waya wa $5,000."}
|
||||
{"id": "test-sw-001", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Kwa hiyo, watu wazima hawana haja ya kufundisha watoto wa shule ya kwanza katika kujifanya, kama wanavyofanya wakati wa kuwasaidia puzzles au kazi nyingine zinazofanana."}
|
||||
{"id": "test-sw-002", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Ni nafasi yetu pekee... Maneno yake mengine yalipotelea kwa kelele ya mikono iliyosisitiza ya kwamba msichana yule ashikwe mateka."}
|
||||
{"id": "test-sw-003", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Kwa upande mwingine, kuna majukumu kama vile mipangilio ya IT na usimamizi ambao lazima uwe ndani ya kampuni."}
|
||||
{"id": "test-th-000", "split": "test", "source_label": "th", "gold_label": "ot", "text": "ชายหาดคือสถานที่สวยงานและเป็นสถานที่ที่ดีที่จะไปดังนั้นนั่นอาจจะเป็นหนึ่งในสถานที่โปรดปรานที่สุดของฉันที่จะไป แล้วคุณล่ะ"}
|
||||
{"id": "test-th-001", "split": "test", "source_label": "th", "gold_label": "ot", "text": "เอ่อ อย่างไรก็ตามฉันสงสัยพลเมืองในมาดริดและแอตแลนตา ในขณะที่พวกเขาอาจเสียดายการสูญเสียบางส่วนของประเพณี แล้วชอบความทันสมัย"}
|
||||
{"id": "test-th-002", "split": "test", "source_label": "th", "gold_label": "ot", "text": "บ่อยครั้งที่คนเดียวที่สามารถรักษา caida de mollera เป็น curandera"}
|
||||
{"id": "test-th-003", "split": "test", "source_label": "th", "gold_label": "ot", "text": "Shakur ได้รับการยืนยันตัวโดยเจ้าหน้าที่ของสเปนในฐานะ Farid Hilali"}
|
||||
{"id": "test-tr-000", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "New Republic’in Charles Lane’i, kaçırma haberinin sadece Gabriel Garcaa Marquez’in dürüst olmayan gazetecilik kayıtlarını uzattığını söylüyor."}
|
||||
{"id": "test-tr-001", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "New Republic’in Charles Lane’i, kaçırma haberinin sadece Gabriel Garcaa Marquez’in dürüst olmayan gazetecilik kayıtlarını uzattığını söylüyor."}
|
||||
{"id": "test-tr-002", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "Yüzbaşı, dedi ve konuşurken, takip eden gemilere işaret etti, Albay Bishop bizi tutuyor."}
|
||||
{"id": "test-tr-003", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "Savaştan sonra yapılan bir örnek evlerdeki geliştirmelerde en çok bulunan şeylerden biri de çocuklardı ve çocukların şehri olarak düşünüldüğünde çok iyi tasarlanmışlardı."}
|
||||
{"id": "test-ur-000", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "تو ویسے ہی، والد چلا جاتا ہے اور میرے لئے چاکلیٹ دودھ کا یہ اچھا بڑا گلاس بنا دیتا ہے."}
|
||||
{"id": "test-ur-001", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "مشورے اور عدالت یا انتظامی ایجنسی کی مخالفت کرنے کے لئے نوٹس بھیجا جائیں گے."}
|
||||
{"id": "test-ur-002", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "انٹیلی جنس طور پر، یہ ممکن نہیں ہوتا کہ غیر مساوی پیچیدہ اداروں کے اس سیارے بڑے بندوق سے بڑے پیمانے پر پیدا ہوسکتے ہیں."}
|
||||
{"id": "test-ur-003", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "دیکھو تم بڑی مصیبت میں آگۓ ہو"}
|
||||
{"id": "test-vi-000", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Người câu cá nước ngọt phải có giấy phép, hãy hỏi văn phòng du lịch gần nhất để biết thông tin về cách lấy giấy phép."}
|
||||
{"id": "test-vi-001", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Đây là kết quả mới nhất: 5,615 nhóm đối tượng tham gia hội cựu sinh viên không đóng góp, 81 1.4 phần trăm, người đóng góp nhiều nhất là $2,840 và người đóng góp ít nhất là $5."}
|
||||
{"id": "test-vi-002", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Nghiệp vụ không chuyển đổi-- được và mất"}
|
||||
{"id": "test-vi-003", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Câu hỏi duy nhất về cuộc khảo sát NEA là Bạn đã đọc bất kỳ tài liệu nào trong năm qua chưa?"}
|
||||
{"id": "test-zh-000", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "根本不是那么回事，用指甲都能划出痕迹 用钥匙那叫一个惨不忍睹啊"}
|
||||
{"id": "test-zh-001", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "邮寄地址错误，请尽快处理，上次回复说是调查清楚后给我回电，又过去一个礼拜，一点消息没有！"}
|
||||
{"id": "test-zh-002", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "面料很好，透气舒服，但是档部设计不好，穿着很难受，不像是可以直接穿的内裤"}
|
||||
{"id": "test-zh-003", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "我订购的是独家签名版，为何收到的书是无签名的？？"}
|
||||
+160
@@ -0,0 +1,160 @@
|
||||
{"id": "train-ar-000", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "اسفل على اليمين مطعم ماكسيم , الذي بدا ك a صالون ايس كريم وهو الان a نصب اكثر الموقر من مادلين ."}
|
||||
{"id": "train-ar-001", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "هل تحب الافلام المعلقة او هل تحب فقط اكشن او"}
|
||||
{"id": "train-ar-002", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "اعاد الموقع فورا موقع برادلي الرسمي , موقع لاخبار عن السباق الرئاسي لعام 2000 , معلومات عن البطاقة الطبية , والكتب عن برادلي ."}
|
||||
{"id": "train-ar-003", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "هل انت مشترك في اي هندسة رسم الاشياء التي"}
|
||||
{"id": "train-ar-004", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "00-187-الاثار المترتبة على الضمان الاجتماعي بالنسبة للمعاشات التقاعدية الخاصة ( جاو / hehs-00-187 , 14 ايلول / سبتمبر 2000 ) ."}
|
||||
{"id": "train-ar-005", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "ويشمل موقع anao وصلات بمنشوراتها المختلفة , بما في ذلك تقارير مراجعة الحسابات و يهدي الممارسات الافضل ."}
|
||||
{"id": "train-ar-006", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "شاليت يقول انه عندما تمشي في الشارع يمكنك ان تخبر العذارى بسبب وهجهم الطازج والمتوازن ."}
|
||||
{"id": "train-ar-007", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "- ويكاد يكون تصنيع حافزا حافزا مكرسا تماما لتطبيقات توليد الطاقة ."}
|
||||
{"id": "train-bg-000", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "това е на описание на г-н браун !"}
|
||||
{"id": "train-bg-001", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "за финансовата 2002 година vba преразгледа плановете си за ефективност на висшите ръководители в регионалните служби за подобряване на индивидуалната отчетност за елементите на изпълнение чрез свързване на целите на организацията за ефективност и действителните резултати с значими и измерими елементи на ефективност ."}
|
||||
{"id": "train-bg-002", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "журналистът е измамник , предназначен само да кара трафика до порно сайтовете на собственика ."}
|
||||
{"id": "train-bg-003", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "толкова добре , колкото хората казват , че е ."}
|
||||
{"id": "train-bg-004", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "о , имаш ли такова време в северна каролина ?"}
|
||||
{"id": "train-bg-005", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "както и да е , закуската е била принуден в тях сред жените преди известно време , така че няма за какво да се тревожим ."}
|
||||
{"id": "train-bg-006", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "тийнейджърите биха искали да се присъединят към на диско орди в хараджуку , близо до парка yoyogi ."}
|
||||
{"id": "train-bg-007", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "циници стенеше , че хитчинс се е цел в клинтън , но вместо това е застрелял блументал ."}
|
||||
{"id": "train-de-000", "split": "train", "source_label": "de", "gold_label": "de", "text": "Sieht gut aus und hat super an meinen skoda Schlüssel gepasst"}
|
||||
{"id": "train-de-001", "split": "train", "source_label": "de", "gold_label": "de", "text": "Kann mich obenstehenden Rezensionen nur anschließen, das Headset ist im Grunde genommen sein Geld wert, würde es nicht nach einem Jahr mit Wackelkontakt den Geist aufgeben."}
|
||||
{"id": "train-de-002", "split": "train", "source_label": "de", "gold_label": "de", "text": "Eine Flasche ist leider mehr oder weniger ausgelaufen. Nichtsdestotrotz tut die Kochsalzlösung was sie soll. Punktabzug gibt es nur wegen dem auslaufen."}
|
||||
{"id": "train-de-003", "split": "train", "source_label": "de", "gold_label": "de", "text": "Habe die Tasche als Geschenk gekauft - Lieferung funktionierte spitze und die Tasche sieht wirklich toll aus (Verarbeitung, Stoff usw.! Daumen hoch!"}
|
||||
{"id": "train-de-004", "split": "train", "source_label": "de", "gold_label": "de", "text": "Gut finde ich die Optik und die Ersatzbürsten. Nicht ganz durchdacht finde ich den Behälter, wo die Klobürste rein kommt. Nimmt man diese raus und steckt sie wieder ein, reibt der Behälter über den Fliesenboden (den man üblicherweise in der Toilette hat) was was Geräuche macht und einen schlechten Qualitätseindruck hinterlässt. Sinnvoll wäre, wenn auf der Unterseite noch eine Gummifläche geklebt würde, so dass Metall nicht auf Fliesen reibt."}
|
||||
{"id": "train-de-005", "split": "train", "source_label": "de", "gold_label": "de", "text": "Der Hut kam sehr spät und fällt ziemlich klein aus. Er sieht aber gut aus und kann praktisch zusammen geklappt werden. Für den Preis - alles in Ordnung."}
|
||||
{"id": "train-de-006", "split": "train", "source_label": "de", "gold_label": "de", "text": "Das Kleidchen ist wirklich schön. Allerdings wirklich nur was für den Strand. Ich trage normalerweise eine S. Hab mir eine M bestellt, damit es etwas lockerer sitzt, aber ich würde sagen, dass sogar eine L angebracht wäre."}
|
||||
{"id": "train-de-007", "split": "train", "source_label": "de", "gold_label": "de", "text": "Ich hatte diese Modell vor fünf Jahren gekauft. War damals super Qualität und hat auch lange gehalten. Jetzt das gleiche Modell von HAMA hier nachgekauft und große Enttäuschung. Kabel viel dünner als vor fünf Jahren. Verarbeitung viel schlechter. Nur günstige Materialien verwendet. Habe das Headset nun schon zum zweiten Mal umgetauscht und eine neues erhalten. Im täglichen Gebrauch hält es etwa zwei Wochen, dann schauen überall die Drähte raus und die Plastikteile lösen sich. Leider nicht mehr zu empfehlen !!!"}
|
||||
{"id": "train-el-000", "split": "train", "source_label": "el", "gold_label": "el", "text": "ΠΊΝΑΚΑΣ 1 : σύγκριση των ηγετικών πρακτικών και των ομοσπονδιακών πρακτικών διαχείρισης των cio"}
|
||||
{"id": "train-el-001", "split": "train", "source_label": "el", "gold_label": "el", "text": "Η ευελιξία του προϋπολογισμού μειώνεται δραστικά , έτσι ώστε μέχρι το 2050 , το καθαρό ενδιαφέρον για το χρέος θα απορροφήσει περίπου το ήμισυ όλων των ομοσπονδιακών εσόδων ."}
|
||||
{"id": "train-el-002", "split": "train", "source_label": "el", "gold_label": "el", "text": "Για όλο το ιστορικό μεγαλείο της , η πιστή γερουσία τώρα είναι το ισοδύναμο ενός Δημοτικού Συμβουλίου , η πολιτική του ικανότητα αφιερωμένη στις προμήθειες νερού , τις γραμμές αποχέτευσης και την ίδρυση παιδότοπους ."}
|
||||
{"id": "train-el-003", "split": "train", "source_label": "el", "gold_label": "el", "text": "Θα πρέπει να πληρούνται οι απαιτήσεις για την κατάρτιση , την τεκμηρίωση και τη συντήρηση ."}
|
||||
{"id": "train-el-004", "split": "train", "source_label": "el", "gold_label": "el", "text": "Προστατευμένη από τους κρύο , υγρασία , βορειοδυτικά ανέμους από τα βουνά των vosges , οι αμπελώνες της αλσατία απολαμβάνουν ένα ιδανικό μικροκλίμα για την παραγωγή λευκών κρασιών που έχουν την εμπιστοσύνη τους στα πιο διάσημα κρασιά της βουργουνδίας και του μπορντώ ."}
|
||||
{"id": "train-el-005", "split": "train", "source_label": "el", "gold_label": "el", "text": "Και κάναμε πολλή κηπουρική εκεί έξω και κυρίως σε μεγάλωσε κρεβάτια για να κρατήσουμε το χώμα ωραίο ."}
|
||||
{"id": "train-el-006", "split": "train", "source_label": "el", "gold_label": "el", "text": "Αχ μου αρέσει η γυναίκα μου δεν μπορεί να καταλάβει ότι θα είναι εκατό βαθμούς έξω θα είμαι εκεί έξω αλλά κάνει πολύ ζέστη για να δουλεύω στην αυλή είμαι κάτω από τα δέντρα περνάω καλά και σταματάω να πιω νερό οπότε τι είναι το πρόβλημα ."}
|
||||
{"id": "train-el-007", "split": "train", "source_label": "el", "gold_label": "el", "text": "Απέναντι από το ο . Στο Audoen , δύο ενδιαφέροντες δρόμοι τρέχουν από την οδό Χάι ."}
|
||||
{"id": "train-en-000", "split": "train", "source_label": "en", "gold_label": "en", "text": "This was ordered as a pack of 2 but I only received 1. Emailed the seller but never received a reply. Feeling jipped."}
|
||||
{"id": "train-en-001", "split": "train", "source_label": "en", "gold_label": "en", "text": "Product received in damaged condition. Realized extent of damage after assembly almost completed since came in components. Unable to contact for product return."}
|
||||
{"id": "train-en-002", "split": "train", "source_label": "en", "gold_label": "en", "text": "Much cheaper quality than expected but item did arrive very fast and not damaged."}
|
||||
{"id": "train-en-003", "split": "train", "source_label": "en", "gold_label": "en", "text": "The book started out well and then just went on and on and on....... and never quite got to the point until I just quit reading it. I got to about 60% of the way and simply could not continue. I see where others have simply skipped ahead full chapters to get to the last 15% of the book. By the time I gave up my interest had simply disappeared. Too bad, could have been written in a much more concise and interesting format."}
|
||||
{"id": "train-en-004", "split": "train", "source_label": "en", "gold_label": "en", "text": "Looked like it was going to be another homage to 80's horror in the vein of Ti West but it just doesn't know what it wants to be and wraps everything up in a really sloppy way. Setup, location and atmosphere are all there they just didn't know what to do with all of that after they jumped into the actual story. Kind of a disappointing miss since that is my only real complaint. All of the pieces were there for a great throwback horror movie and it was all squandered"}
|
||||
{"id": "train-en-005", "split": "train", "source_label": "en", "gold_label": "en", "text": "Perfect for adding light layers of hydration especially in the colder months. This is usually my first layer of hydration before a moisturizer and it works great with no pilling or balling. Hydration lasts all day!"}
|
||||
{"id": "train-en-006", "split": "train", "source_label": "en", "gold_label": "en", "text": "Very big and gaudy looking. Will not wear them."}
|
||||
{"id": "train-en-007", "split": "train", "source_label": "en", "gold_label": "en", "text": "Product description says gown but its an all in one pants and way top big for a newborn. The headband stitching is coming undone, very disappointed."}
|
||||
{"id": "train-es-000", "split": "train", "source_label": "es", "gold_label": "es", "text": "Por lo que vale y con premium te sacas la compra del mes. Trae funda y dos protectores por un gran precio."}
|
||||
{"id": "train-es-001", "split": "train", "source_label": "es", "gold_label": "es", "text": "Son finas de algodón 100%, ideales para el veranito, yo las he comprado para regalar a una amiga que ha dado a luz hace un par de meses."}
|
||||
{"id": "train-es-002", "split": "train", "source_label": "es", "gold_label": "es", "text": "Buenísimos. Pintan genial. Los he probado sobre cristal y tienen una solidez que permite que se vean perfectamente. Se mantienen durante mucho tiempo pero a la vez de borran con facilidad. Los recomiendo 100%. De hecho será la marca que compre a partir de ahora."}
|
||||
{"id": "train-es-003", "split": "train", "source_label": "es", "gold_label": "es", "text": "Me ha gustado mucho tanto la calidad de los materiales como los acabados que tiene, el toque de la madera es muy chulo. Tiene las clásicas tres velocidades aunque como ya sabéis se usa en el 90% de las veces la misma. Se puede colocar unos altavoces por la parte de atrás, la verdad que suena bastante más alto de lo que me pensaba, aún así he conectado los dos altavoces que tengo repartidos por el comedor y el resultado es espectacular. Unos recuerdos del sonido clásico que me ha gustado mucho recordar, estoy muy contento con el resultado. A destacar la superficie de la maleta, estéticamente es muy bonita, incluso simplemente para tener colocado encima de algún mueble, cumple perfectamente los dos cometidos, la estética y la funcional. Por supuesto cuenta con entrada USB y con función bluetooth, perfecto porque puedes reproducir también la música del teléfono móvil."}
|
||||
{"id": "train-es-004", "split": "train", "source_label": "es", "gold_label": "es", "text": "Muy robusto, era para niño de 8 años pero como es regulable le servirá para mucho"}
|
||||
{"id": "train-es-005", "split": "train", "source_label": "es", "gold_label": "es", "text": "Me gusta brillo de labios, es muy poca cantidad pero esta bien."}
|
||||
{"id": "train-es-006", "split": "train", "source_label": "es", "gold_label": "es", "text": "No es compatible para el nuevo XS, todas las fundas son del X y las venden como XS y no saben que no es igual, solo hay que ver el altavoz de al rededor de la cámara"}
|
||||
{"id": "train-es-007", "split": "train", "source_label": "es", "gold_label": "es", "text": "Esta bien pero pense que incluía las bombillas..."}
|
||||
{"id": "train-fr-000", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Bien reçu bien emballer fonctionne correctement, j’avais peur du bruit par rapport au commentaires mais le bruit des ventilateurs est correct"}
|
||||
{"id": "train-fr-001", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Ce produit est tout simplement génial. il agit très vite. l'odeur disparait quasiment au moment de la pulvérisation, il n'y a pas pas besoin d'en mettre beaucoup le flacon dure longtemps. Ensuite le nettoyage est très facile et il y a même comme une petite odeur de frais. Le chat ne reviens pas faire au même endroit après avoir nettoyé avec ce produit. N'hésitez pas."}
|
||||
{"id": "train-fr-002", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Beau mais pour coller c'est pas très efficace. J'en ai essayé 2 ils n'ont pas tenu 1 semaine. Mais j'en ai fixé un avec une vis. Donc on peut l'adapter."}
|
||||
{"id": "train-fr-003", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "L' idée m' avait séduit mais l' objet est décevant. Les chiffres sont de simples autocollants , la rotation des pièces n' est pas fluide. Je n' aurais pas acheté si j' avais pu le toucher."}
|
||||
{"id": "train-fr-004", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Torche puissante, idéale pour un vélo, le système de fixation semble fiable. En revanche, je regrette la batterie fournie et nécessaire qui ne correspond à aucune classe de pile classique."}
|
||||
{"id": "train-fr-005", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "LIVRAISON RAPIDE ET SERIEUSE"}
|
||||
{"id": "train-fr-006", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Je viens de m'apercevoir que ce produit vendu comme neuf semble être reconditionné puisque une photo est restée dans l'appareil (personnes et lieu que je ne connais pas du tout) . La date de la photo est antérieure à ma date d'achat."}
|
||||
{"id": "train-fr-007", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Simple jouet avec lequel on s'amuse deux minutes avec , la mode est finis"}
|
||||
{"id": "train-hi-000", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "हम प ् रत ् येक राज ् य के लिए कर कोड और प ् रक ् रियाओं को प ् राप ् त कर सकते हैं , इन ् हें जांच करें , चुनिंदा अधिकारियों का साक ् षात ् कार करें और कुछ विश ् वसनीय पैटर ् न उत ् पन ् न करें ."}
|
||||
{"id": "train-hi-001", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "और उम हाँ उह हाँ मुझे लगता है कि मैं आमतौर पर मैं हूँ उम भारी सॉस और उम"}
|
||||
{"id": "train-hi-002", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "जून में , जो कि जून में अपेक ् षित है , वह व ् हाइट हाउस वकील के जनादेश के न ् यायालय की व ् याख ् या को चालू करेगा . इस निर ् णय का कोई कानूनी उदाहरण नहीं है ."}
|
||||
{"id": "train-hi-003", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "Anao की साइट में उनके विभिन ् न प ् रकाशनों के लिंक शामिल हैं , जिसमें अपनी लेखापरीक ् षा रिपोर ् ट और बेहतर अभ ् यास गाइड शामिल हैं ."}
|
||||
{"id": "train-hi-004", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "11 वीं राजवंश की अवधि से पीटर के प ् राचीन मिस ् री धर ् म , सी . 2134 ईसा पूर ् व"}
|
||||
{"id": "train-hi-005", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "हाँ हाँ किसी ने कहा कि उम कि इस तरह के कुछ प ् रकार के शुरू में कुछ प ् रकार की है और यह कि जब कार ठंडा हो जाता है , तो धातु को छोटा बनाने के लिए फैलता है और तब तक यह नहीं है कि कार को नीचे ठंडा होने तक नहीं है धातु आप जानते हैं कि इन दो टुकड ़ े किसी और को छू नहीं रहे हैं और आप और आप जानते हैं कि आप जानते हैं तो यह शुरू हो जाएगा क ् योंकि आपके पास छोटा नहीं है"}
|
||||
{"id": "train-hi-006", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "लेखा विषयों के अतिरिक ् त , निर ् देशिका सूची में कुछ ऐसी एजेंसियों या प ् रोग ् राम जो उदाहरण में प ् रयोग किया गया है या उसके मानकों के भीतर अद ् वितीय प ् रावधान है ."}
|
||||
{"id": "train-hi-007", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "वे चीज ़ ें जो केवल आप को प ् रभावित कर रहे हैं , उम कि हम लोगों को कैसे दंडित करते हैं और क ् यों चीजें हैं जैसे कि वे रास ् ते हैं"}
|
||||
{"id": "train-it-000", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Gli aerei da guerra russi colpiscono in Siria"}
|
||||
{"id": "train-it-001", "split": "train", "source_label": "it", "gold_label": "ot", "text": "L'uomo sta portando una cassetta degli attrezzi sul marciapiede."}
|
||||
{"id": "train-it-002", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Israele trattiene 37 palestinesi mentre prosegue l'operazione di arresto"}
|
||||
{"id": "train-it-003", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Uomo ucciso in un raid del terrore francese"}
|
||||
{"id": "train-it-004", "split": "train", "source_label": "it", "gold_label": "ot", "text": "La gente va e pagaia su una zattera."}
|
||||
{"id": "train-it-005", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Mi oppongo alla pena di morte."}
|
||||
{"id": "train-it-006", "split": "train", "source_label": "it", "gold_label": "ot", "text": "La donna sta condendo l'olio."}
|
||||
{"id": "train-it-007", "split": "train", "source_label": "it", "gold_label": "ot", "text": "In una dichiarazione via e-mail al Knoxville News Sentinel, Shumaker ha detto: \"Non prendo in considerazione le dimissioni."}
|
||||
{"id": "train-ja-000", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "味はとても良かったのですが、もうちょっと手頃な価格だと良かったのですが・・・"}
|
||||
{"id": "train-ja-001", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "ジュースに入れて使用しています。まだ加熱料理にしようしていませんが、機会があれば使用してみたいと思います。"}
|
||||
{"id": "train-ja-002", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "７キロ用の洗濯機にはパツンパツンです。表記のサイズはいい加減なので、１〜２cm小さいと考えてください。シャープのプラズマクラスター付きの現行品にテラスで使用のため、このカバーを買いましたが強風で上方にズレるため、引っ張って下に降ろす際に、すでに裏の結び紐の片側が切れてしまいました。簡単に切れるので気を付けたいですね？！ なので価格のことを考えると、６ヶ月（新年を迎えられれば）御の字ですが、果たして３ヶ月持つかどうか？様子見です。前のものがLEC（レック）という老舗のブランド（上場企業なのにアフターの対応も最悪なので２度とこのメーカーの製品は使わないと決めています）のもので１２００円位した割りには、１５ヶ月位しか持たなかったので、３〜４ヶ月使えればと。 →結果として１ヶ月でボロボロ。既に廃棄しました。"}
|
||||
{"id": "train-ja-003", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "一月過ぎたけどまだ来ない。最悪‼️流石中国星1つ付けたけど本当は星-5"}
|
||||
{"id": "train-ja-004", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "埃に気を付け、湿度のある浴室で貼り付けを行いました。埃は問題が無かったのですが、残念ながら他の方も書かれていますが端の方に気泡が残ります、ガラス面と黒い額縁の境目に残ります。 爪で押し出しましたが気泡を追い出せません。 少し剥がし、気泡を追い出しましたがそういう問題で気泡が残っているわけでは無いようで、若干浮いていて気泡が入っている様です。 穴の部分はぴったりの位置にあるのでRの部分が相当シビアに作られているのかも知れません。 うまく貼れた場合でも、使用するとガラス面に結構指紋が付きます、何も貼らなかった場合に比べて滑らかさと防汚の面が劣りますが、これは全てのこういった商品に当てはまるのかも知れません。 ガラスの傷つきにくさですが、何も貼らなかった場合に比べて貼った方が強度は上のようです。 1ヵ月、XZ1をそのまま使用していると光に当てないと分からない位ですが、小傷が沢山出来ていました、こちらの商品を貼って1週間色々ハードに使いましたが傷は出来ておりません、強度は申し分ありません。"}
|
||||
{"id": "train-ja-005", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "マツエク休憩中で数年ぶりにマスカラを買いました。ダマにならず、しっかり付くのに本当にお湯でスルッと取れるので良いです！"}
|
||||
{"id": "train-ja-006", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "固定ができない･･･なんか斜めになる ズームするために回したら望遠本体まで動くから固定できず写真が撮れない。あと三脚なら固定できるだろうと思ってやったらできなかった…。魚眼レンズとかは全然使えるけど1番使いたかった望遠レンズが使えなくて残念。"}
|
||||
{"id": "train-ja-007", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "ずっと前から欲しかったですが、高価なイメージがあって。今回、充電の持ちと手頃なお値段で、こちらを選びました！ペアリングもスムーズにできたし、軽いし、耳へのフィット感も音もすごくいいです。ちなみに電子レンジ使う時だけ（特にレンジの使い始め）たまに途切れます。しかし、全然許容範囲です。"}
|
||||
{"id": "train-nl-000", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Een hond die op een blikje bijt."}
|
||||
{"id": "train-nl-001", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Rusland zegt dat ballistische raketten afgevuurd zijn in het Middellandse Zeegebied..."}
|
||||
{"id": "train-nl-002", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Iran atoombusbesprekingen beginnen in de hoop op vooruitgang..."}
|
||||
{"id": "train-nl-003", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Twee vrouwen die op de bruine bank zitten."}
|
||||
{"id": "train-nl-004", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Er loopt een hond in de sneeuw."}
|
||||
{"id": "train-nl-005", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Freddie Starr gearresteerd in Savile Abuse Sonde..."}
|
||||
{"id": "train-nl-006", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Een hond in een auto."}
|
||||
{"id": "train-nl-007", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Een aanrecht en aanrechtblad met schalen op de planken."}
|
||||
{"id": "train-pl-000", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Dwie dziewczyny z kucykami jeżdżą na przejażdżce w parku rozrywki."}
|
||||
{"id": "train-pl-001", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "alstom konkuruje o kontrakt z krajami japońskimi i niemieckimi."}
|
||||
{"id": "train-pl-002", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Pankda je bambus."}
|
||||
{"id": "train-pl-003", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Libijczycy zaczynają oddawać broń"}
|
||||
{"id": "train-pl-004", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "4 zagranicznych żołnierzy zabitych na wschodzie Afganistanu"}
|
||||
{"id": "train-pl-005", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Egipt głosuje nad nową konstytucją"}
|
||||
{"id": "train-pl-006", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Dwie osoby grają w golfa na polu golfowym."}
|
||||
{"id": "train-pl-007", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Nigdy się z tego nie wydostaliśmy!"}
|
||||
{"id": "train-pt-000", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Combatentes rebeldes 'capturam' soldados sírios"}
|
||||
{"id": "train-pt-001", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Nenhum acordo sobre penhasco fiscal enquanto Obama vai de férias"}
|
||||
{"id": "train-pt-002", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Não foi possível contactar imediatamente representantes seqüenciais para comentários sobre o anúncio da SCO."}
|
||||
{"id": "train-pt-003", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Quatro aviões azuis e amarelos sobrevoando quatro barcos."}
|
||||
{"id": "train-pt-004", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "O Comité Bancário do Senado está agendado para realizar uma audiência na terça-feira, onde Donaldson está agendado para testemunhar sobre hedge e fundos mútuos."}
|
||||
{"id": "train-pt-005", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Thomas Cook acusado de colocar os custos à frente dos clientes"}
|
||||
{"id": "train-pt-006", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Um cão corre à volta de um quintal."}
|
||||
{"id": "train-pt-007", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Um homem está a saltar uma parede."}
|
||||
{"id": "train-ru-000", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Nsiad использует выводы таким образом в рамках своей текущей работы по двусторонним инициативам ."}
|
||||
{"id": "train-ru-001", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Но это также о том времени , когда ты начинаешь получать температуру в кабине ."}
|
||||
{"id": "train-ru-002", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Поощряется проведение экспериментов , равно как и представление такой дополнительной информации , что будет способствовать укреплению финансового доклада ."}
|
||||
{"id": "train-ru-003", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Да , это совсем другая культура . Это странно внизу , потому что"}
|
||||
{"id": "train-ru-004", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Присяжные посмотрели вверх , заинтересовались ."}
|
||||
{"id": "train-ru-005", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Но может ли он навсегда ?"}
|
||||
{"id": "train-ru-006", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "И я могу назвать несколько больше , я просто не могу придумать их имена прямо с рук ."}
|
||||
{"id": "train-ru-007", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Межучрежденческой совет по делам бездомных находится на ранних этапах своего 10-летнего плана по ликвидации бездомности , как это было сказано ."}
|
||||
{"id": "train-sw-000", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Isipokuwa kubwa ni uandishi wa habari wa ufaransa , ambapo heshima wakati kwa mitindo imekuwa ni mtazamo wa kawaida tangu karne ya 17"}
|
||||
{"id": "train-sw-001", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Mipangilio ya uga ."}
|
||||
{"id": "train-sw-002", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Ofisi ya usimamizi wa habari ( oim ) , ofisi ya utendaji wa programu ( opp ) , na timu ya kupanga ya serikali sasa ni wafanyakazi ."}
|
||||
{"id": "train-sw-003", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Nguvu hii ya asili ana na mwanaume tangu mara ya mapema , kama kuonyesha na pango-hekalu la pan , kwa ambaye raia wa syria na greeks kujitolea mkondo huo ."}
|
||||
{"id": "train-sw-004", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Mawasiliano ya kibinafsi ( 2 ) pamoja na ande salimbot , undugu wa kimataifa wa boilermakers , wajenzi wa meli , blacksmiths , wazushi na wasaidizi , februari 22 , 2002 ."}
|
||||
{"id": "train-sw-005", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Watu nani haki zao na kila kitu ng ' ambo na kila kitu na i think uh sijui kwa hiyo"}
|
||||
{"id": "train-sw-006", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Kuunda mazingira ya ushindani wange , kwa kiwango cha chini , kuondoa statutes4 ya kibinafsi na kanuni ya barua pepe ."}
|
||||
{"id": "train-sw-007", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Akiba ya kitaifa ya kitaifa"}
|
||||
{"id": "train-th-000", "split": "train", "source_label": "th", "gold_label": "ot", "text": "โอ้ มัน ฟัง ดู ยอดเยี่ยม"}
|
||||
{"id": "train-th-001", "split": "train", "source_label": "th", "gold_label": "ot", "text": "ขอบใจ หลาย ๆ เด้อ รี เขา จากไป ด้วย การ เดิน แบบ เดียว กับ ที่ เขา มี จาก ซอย ที่ มั่นคง"}
|
||||
{"id": "train-th-002", "split": "train", "source_label": "th", "gold_label": "ot", "text": "แต่ ฉัน เดา ว่า"}
|
||||
{"id": "train-th-003", "split": "train", "source_label": "th", "gold_label": "ot", "text": "ใช่ พวกเขา ควร ทำความสะอาด ให้ ชัดเจน ว่า มัน จะ ไม่ พา พวกเขา มาก ไป ใส่ สแตมป์ บน กระป๋อง น้ำผลไม้ ง่ายๆ เหมือน โซดา"}
|
||||
{"id": "train-th-004", "split": "train", "source_label": "th", "gold_label": "ot", "text": "การ ตั้งครรภ์ ธรรมชาติ ได้รับ การ ยกระดับ โดย ระบบ ของ ยุโรป ที่ ล้ำสมัย ที่สุด ของ ชลบุรี ยังคง ดำเนินการ อยู่ ใน คลอง กลาง ที่ คุณ จะ เห็น บน ทาง ของ คุณ ใต้ ไป ยัง วี"}
|
||||
{"id": "train-th-005", "split": "train", "source_label": "th", "gold_label": "ot", "text": "Atrisk งานก่อสร้าง ที่ แท้จริง มี การแสดง โดย ผู้รับเหมา การค้า ภายใต้ สัญญา กับ ซ. ผู้ แล้ว กลายเป็น ผู้รับผิดชอบ เจ้าของ สำหรับ การ ก่อสร้าง หมายถึง และ วิธีการ และ การจัดส่ง ของ สิ่งอำนวยความสะดวก ที่ สมบูรณ์ ภายใน ขอบเขต ของ เจ้าของงาน สำหรับ ต้นทุน เวลา และ คุณภาพ"}
|
||||
{"id": "train-th-006", "split": "train", "source_label": "th", "gold_label": "ot", "text": "มัน ไม่ รบกวน ฉัน เลย ฉัน ไม่ รู้สึก ว่า มัน เป็นการ ละเมิด ความเป็นส่วนตัว หรือ อะไร เลย"}
|
||||
{"id": "train-th-007", "split": "train", "source_label": "th", "gold_label": "ot", "text": "นี่ คือ ผล งานชิ้นเอก ของ อารยธรรม ฝรั่งเศส ที่ มี รอย เปลี่ยน ใน ศตวรรษ ที่ 12 จาก สไตล์ โร มาน ส ์ แบบ ไม่มีสติ ของ โบสถ์ เริ่ม สู่ ความ แข็งแกร่ง มากขึ้น"}
|
||||
{"id": "train-tr-000", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Ayda 20 zloti veya her ay 20 zloti taksitle uygun bir alışveriş seçeneği sunuyoruz ."}
|
||||
{"id": "train-tr-001", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Amerika ' nın zamanında geleceğini içgüdüsel olarak hissetti ."}
|
||||
{"id": "train-tr-002", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Santorini ve mykonos özellikle avrupa ve ABD ' den bir sürü tasarımcı giyim ve ayakkabı var ."}
|
||||
{"id": "train-tr-003", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Ca ' daan , kara sürücüler karşılığında bir şey söyledi mi diye olabilir ."}
|
||||
{"id": "train-tr-004", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Hareketleri dikkatlice ölçülür ."}
|
||||
{"id": "train-tr-005", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Işte bunu söylediğin için mutluyum ve gerek yoktu ama o da aynı şekilde hissediyorum ben oturup bazı programları izliyorum ve uh ve bile dahil olan insanlar için utanıyorum bunu kendine neden yapıyorsun biliyorsun"}
|
||||
{"id": "train-tr-006", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Kazanan giriş , rock kendini"}
|
||||
{"id": "train-tr-007", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Örneğin , bir bilgisayar sisteminin güvenli olduğu ve tanımlanmış bir ortamda çalışmasına izin veren yazılı bir yetkilendirme ."}
|
||||
{"id": "train-ur-000", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "ماس ماس ( ماس ) نیٹ ورک کی بنیاد پر یہ اثر انداز کیا گیا ہے ."}
|
||||
{"id": "train-ur-001", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "اس میں تاخیر کی جاتی ہے . \""}
|
||||
{"id": "train-ur-002", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "اور ہم نے اس وقت تک ایک مدت معین کیا ہے جس کی وجہ سے اس سے ملاقات کی جاتی ہے تو ہم نے رات کے اوقات میں رات یا اس سے بھی کچھ کم نہیں کیا ۔ کوفتے کی تراکیب"}
|
||||
{"id": "train-ur-003", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "پھر ہم نے وزن کیا اور ہم نے اس میں سے ہر ایک کو عاجز کر دیا اور ہم ان سے بیزار ہیں"}
|
||||
{"id": "train-ur-004", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "کچھ بھی نہیں ۔"}
|
||||
{"id": "train-ur-005", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "جب میں نے آخری بار ایک سال میں پہلی بار ایسوسی ایشن سے اظہار کیا ، جس کی وجہ سے آپ نے ایک دوسرے کے پیمانے پر کام کرنے کی کوشش کی ، اس کے بعد ، دیگر ٹرانسپورٹ فرمیں کے ساتھ کام کرنے کے لئے ."}
|
||||
{"id": "train-ur-006", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "یا آپ کا خیال ہے کہ آپ کو اس طرح کے کم از کم ایک بار نظر آنا چاہتے ہیں ؟"}
|
||||
{"id": "train-ur-007", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "واقعی رئیل اسپورٹس"}
|
||||
{"id": "train-vi-000", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Một ứng cử viên alpha sẽ không cần sói chút nào cả ."}
|
||||
{"id": "train-vi-001", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Cô ấy là một người phụ nữ đáng chú ý , tăng sức mạnh làm nhiếp chính cho những thanh niên tutmosis ii con trai của cô ấy trước khi lấy nó cho chính mình bằng cách tự xưng là quyền thống trị ."}
|
||||
{"id": "train-vi-002", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Những sự kiện đặc biệt như thú cưng cho thấy , thử thách chó , và những ngày thái lịch sử được tổ chức trong suốt mùa hề ."}
|
||||
{"id": "train-vi-003", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Bây giờ hãy nhìn vào số tiền mà chính phủ có thể tiết kiệm nếu họ không có tất cả những ngày nghỉ trong những ngày nghỉ đó ."}
|
||||
{"id": "train-vi-004", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "The Minato Mirai 21 dự án , được khởi chạy vào giữa những năm 1980 , đã được dự định sẽ biến một đường rộng lớn của bờ sông phía bắc và phía đông của sakuragi-cho vào một thành phố mô hình của tương lai , tích hợp kinh doanh , triển lãm , và giải trí cơ sở ."}
|
||||
{"id": "train-vi-005", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Tiêu chuẩn quốc tế một số công ty lớn / công ty e đã bảo vệ 9000 chứng nhận là một tổ chức chất lượng ( ISO ) 9000 hoạt động kiểm soát ."}
|
||||
{"id": "train-vi-006", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Tất cả các bạn sẽ đến ăn tối với tôi ở savoy ."}
|
||||
{"id": "train-vi-007", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Ca ' daan ' không thể thấy được nếu những kỵ sĩ đen nói bất cứ điều gì để trở lại ."}
|
||||
{"id": "train-zh-000", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "东西不错，携带方便。但中国人鼻子比较低的，用的时候容易掉下来。"}
|
||||
{"id": "train-zh-001", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "商品预测还有两周送货，我就提前申请退货取消订单，但是还是送过来了，联系客服怎么处理？客服说海外购订单无法取消，呵呵，买了就不许退，就这么硬气，然后退货必须自己出运费，什么玩意，垃圾商家亚马逊"}
|
||||
{"id": "train-zh-002", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "本书共计368页，前57页包括解释全文，答记者问，理解适用，逐条释义，57页到165页案例，剩下部分都是法条，这么高定价，太黑了，请各位参考购买"}
|
||||
{"id": "train-zh-003", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "今天收到衣服， 很感人的卡片，真是体现的服务。超赞！"}
|
||||
{"id": "train-zh-004", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "2016年4月11日下午4：31收到的天美时手表打开发现手表直接用表带绑在表盒内的塑料支架上，没有任何贴膜等保护措施，表带又轻又软质感很差，随后当天申请退货，4月12日上午顺丰快递寄回客服发的亚马逊钟表/珠宝频道，4月14日接到天津亚马逊打来电话，说经亚马逊检验手表没有贴膜且有磨损不符合退货要求。这就奇怪了 ，为什么寄出的时候根本不存在的贴膜，在退货检验的时候就变成理由了呢；为什么寄出的时候没有任何保护措施且直接绑在一个硬塑料支架上的这种包装方式反倒成了退货时手表有痕迹拒绝退货的理由了；而且网站的包装清单上没有注明任何关于这款手表应该有贴膜信息，如果说新品真的包含贴膜，那只能说明亚马逊将二手货和新品混合出售，这样作为消费者根本无法保护自己的权益；现在就造成亚马逊随便找一个理由就可以把这种二手产品直接推给消费者，这难道不属于变相的欺诈么。在此提醒其他想要购买亚马逊产品的消费者，至少中国亚马逊是没有道德底线的，高价商品千万不能在这购买。4月14日将这个情况反映到亚马逊之后，客服说会备注信息向上反映，4月16日再联系亚马逊时，被告知没有任何人员跟踪此事，亚马逊又开始采取拖延战术了。"}
|
||||
{"id": "train-zh-005", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "清仓的东西包装就这样 盒子发来就是开的还不是我打开的。不过东西不错啊 应该是全新没问题"}
|
||||
{"id": "train-zh-006", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "是我在亚马逊够买最不满意的物品，为什么换了这型号芯，出来的水烧开后有水渍，反而自来水却没有。为什么？厂家能解释吗？"}
|
||||
{"id": "train-zh-007", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "这车的价格太贵了，以为这个价格可以买好点的玩具，一打开包装我也是醉了，有种被骗的心"}
|
||||
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"heldout_rows": 80,
|
||||
"train_test_exact_text_overlap": 0,
|
||||
"teacher_gold_accuracy": 1.0,
|
||||
"baseline_gold_accuracy": 0.0,
|
||||
"trained_gold_accuracy": 0.95,
|
||||
"trained_absolute_uplift": 0.95,
|
||||
"trained_teacher_agreement": 0.95,
|
||||
"teacher_mean_latency_seconds": 4.143897417721019,
|
||||
"trained_mean_latency_seconds": 0.021054333343636246,
|
||||
"latency_speedup": 196.81921769199724,
|
||||
"teacher_input_tokens": 30898,
|
||||
"trained_input_tokens": 7738,
|
||||
"input_token_reduction": 0.7495630785164089,
|
||||
"teacher_provider_cost_usd": 0.13577433662337665,
|
||||
"trained_provider_charge_usd": 0.0,
|
||||
"cost_scope_note": "Local inference made no provider API calls; electricity/hardware amortization is excluded."
|
||||
}
|
||||
+18
@@ -0,0 +1,18 @@
|
||||
{
|
||||
"dataset": "papluca/language-identification",
|
||||
"revision": "aa56583bf2bc52b0565770607d6fc3faebecf9e2",
|
||||
"train_source_split": "train",
|
||||
"test_source_split": "test",
|
||||
"train_fingerprint": "f45582c1a8dc5a98",
|
||||
"test_fingerprint": "24705149585614f9",
|
||||
"seed": 78,
|
||||
"train_per_source_language": 8,
|
||||
"test_per_source_language": 4,
|
||||
"train_rows": 160,
|
||||
"test_rows": 80,
|
||||
"train_unique_texts": 160,
|
||||
"test_unique_texts": 75,
|
||||
"exact_text_overlap": 0,
|
||||
"train_rows_sha256": "47b06a13660737187da7e8dede85daffcb4920a57734c2f8a4b20eaa31d20b31",
|
||||
"test_rows_sha256": "b99d21173e89f8d37de4c4a11a32ae8aca0f5e8de6b384d366114a62afb757c1"
|
||||
}
|
||||
@@ -0,0 +1,134 @@
|
||||
{
|
||||
"schema_version": 1,
|
||||
"experiment_id": "8-8",
|
||||
"status": "complete",
|
||||
"created_at": "2026-08-02T16:55:07.929987+00:00",
|
||||
"command": "/Users/boj/book/ai-agent-book/.venv/bin/python run_experiment_8_8.py --run-dir validation/exp8-8-kimi3-smollm2-20260730 --phase package",
|
||||
"host": {
|
||||
"platform": "darwin",
|
||||
"python": "3.11.4 (main, Jul 5 2023, 08:40:20) [Clang 14.0.6 ]",
|
||||
"credential_env": "MOONSHOT_API_KEY"
|
||||
},
|
||||
"student": {
|
||||
"model": "HuggingFaceTB/SmolLM2-135M-Instruct",
|
||||
"revision": "12fd25f77366fa6b3b4b768ec3050bf629380bac"
|
||||
},
|
||||
"teacher": {
|
||||
"model": "kimi-k3",
|
||||
"base_url": "https://api.moonshot.cn/v1",
|
||||
"pricing": {
|
||||
"input_per_million": 20.0,
|
||||
"cached_input_per_million": 2.0,
|
||||
"output_per_million": 100.0,
|
||||
"currency": "CNY",
|
||||
"source_url": "https://platform.kimi.com/docs/pricing/chat-k3.md",
|
||||
"as_of": "2026-07-29",
|
||||
"usd_per_currency_unit": 0.1477922077922078,
|
||||
"fx_source_url": "https://www.ecb.europa.eu/stats/eurofxref/eurofxref-daily.xml",
|
||||
"fx_as_of": "2026-07-29"
|
||||
}
|
||||
},
|
||||
"gates": {
|
||||
"disjoint_heldout": true,
|
||||
"real_teacher_receipts": true,
|
||||
"real_parameter_training": true,
|
||||
"before_after_quality": true,
|
||||
"quality_near_teacher": true,
|
||||
"measured_latency": true,
|
||||
"measured_tokens": true,
|
||||
"dollar_accounting": true
|
||||
},
|
||||
"comparison": {
|
||||
"heldout_rows": 80,
|
||||
"train_test_exact_text_overlap": 0,
|
||||
"teacher_gold_accuracy": 1.0,
|
||||
"baseline_gold_accuracy": 0.0,
|
||||
"trained_gold_accuracy": 0.95,
|
||||
"trained_absolute_uplift": 0.95,
|
||||
"trained_teacher_agreement": 0.95,
|
||||
"teacher_mean_latency_seconds": 4.143897417721019,
|
||||
"trained_mean_latency_seconds": 0.021054333343636246,
|
||||
"latency_speedup": 196.81921769199724,
|
||||
"teacher_input_tokens": 30898,
|
||||
"trained_input_tokens": 7738,
|
||||
"input_token_reduction": 0.7495630785164089,
|
||||
"teacher_provider_cost_usd": 0.13577433662337665,
|
||||
"trained_provider_charge_usd": 0.0,
|
||||
"cost_scope_note": "Local inference made no provider API calls; electricity/hardware amortization is excluded."
|
||||
},
|
||||
"artifacts": {
|
||||
"benchmark_test_gold.jsonl": {
|
||||
"sha256": "8ae7f2966c7f955b76af51f7996a0ed626eb7eb595fdc8f594e4b46ec960b33f",
|
||||
"bytes": 19022
|
||||
},
|
||||
"benchmark_train_gold.jsonl": {
|
||||
"sha256": "fe617c0fbe0344685976f95cef7335a77c17f526381927b9f3af321276a298fb",
|
||||
"bytes": 45810
|
||||
},
|
||||
"comparison.json": {
|
||||
"sha256": "baad4abd94d3284125b1bf148e88c12f9e70dc5f7125f00f87b8c93b235b6659",
|
||||
"bytes": 700
|
||||
},
|
||||
"dataset_provenance.json": {
|
||||
"sha256": "ab9d83cefc1bcf019707a454593deaf98ca97b9101518648345e7c87bec9a6ba",
|
||||
"bytes": 643
|
||||
},
|
||||
"student_adapter/README.md": {
|
||||
"sha256": "f9b48a87d653b2af6971a2240d29ab5a8929fe7e156cc566102f50da655aefa2",
|
||||
"bytes": 5224
|
||||
},
|
||||
"student_adapter/adapter_config.json": {
|
||||
"sha256": "1e72f90088e3c7a25d624b6fee8e5275f3d7b67be3d4664e44e0d3e8d010135a",
|
||||
"bytes": 1036
|
||||
},
|
||||
"student_adapter/adapter_model.safetensors": {
|
||||
"sha256": "88e7e5844b34cc2f14bded556bcbd19fb8b532d90c3db6c0772bdc9350257e25",
|
||||
"bytes": 3702168
|
||||
},
|
||||
"student_adapter/chat_template.jinja": {
|
||||
"sha256": "872be49dbb638044ad01b60388f48d469ff2980e5f0dccdc22ec907db54d0788",
|
||||
"bytes": 368
|
||||
},
|
||||
"student_adapter/tokenizer.json": {
|
||||
"sha256": "bf346d64f6f0fbcefb4c1b6928a98241467dff36c6fbae5fe1785c4ff90667f4",
|
||||
"bytes": 3522871
|
||||
},
|
||||
"student_adapter/tokenizer_config.json": {
|
||||
"sha256": "3e9c0d0aff40796f26a2986e6ed1b5e03646765272eb20bacd2492bd020e6c98",
|
||||
"bytes": 453
|
||||
},
|
||||
"student_baseline_heldout.jsonl": {
|
||||
"sha256": "20ed17a02e0847adc2cf4b6d54928a43f166adf7901f0717dd43241684413b8e",
|
||||
"bytes": 17363
|
||||
},
|
||||
"student_baseline_summary.json": {
|
||||
"sha256": "4c5984e0e84972bcdc8996029275f882274c171fc2f60d9f69783efc765883dd",
|
||||
"bytes": 434
|
||||
},
|
||||
"student_train.jsonl": {
|
||||
"sha256": "2e79dacebb153c68d0509e9106b8f194d97a157143194c85f2399eaafde97306",
|
||||
"bytes": 61010
|
||||
},
|
||||
"student_trained_heldout.jsonl": {
|
||||
"sha256": "1f35f16f4b0be3f3790483d71bff743ee523c84a6a52d4f1e83c97cefeea2f6d",
|
||||
"bytes": 16005
|
||||
},
|
||||
"student_trained_summary.json": {
|
||||
"sha256": "74274043a580c902e8267069e7307a9bf092032858f11d6385838767fcc3d698",
|
||||
"bytes": 437
|
||||
},
|
||||
"teacher_receipts.jsonl": {
|
||||
"sha256": "133440abe4fb41905b1fceff46e561ff212521b4530c0917480c96aa49c16119",
|
||||
"bytes": 343657
|
||||
},
|
||||
"teacher_summary.json": {
|
||||
"sha256": "12e7ed9fc21a531f5e8817f30dc664d28ebe137170cdd20dda9b7de0e1960788",
|
||||
"bytes": 1457
|
||||
},
|
||||
"training_receipt.json": {
|
||||
"sha256": "ac07644a977ee45c0985f52fb013388375f1d0b126ff19de773f691260fe8e3e",
|
||||
"bytes": 5824
|
||||
}
|
||||
},
|
||||
"credential_values_retained": false
|
||||
}
|
||||
+207
@@ -0,0 +1,207 @@
|
||||
---
|
||||
base_model: HuggingFaceTB/SmolLM2-135M-Instruct
|
||||
library_name: peft
|
||||
pipeline_tag: text-generation
|
||||
tags:
|
||||
- base_model:adapter:HuggingFaceTB/SmolLM2-135M-Instruct
|
||||
- lora
|
||||
- transformers
|
||||
---
|
||||
|
||||
# Model Card for Model ID
|
||||
|
||||
<!-- Provide a quick summary of what the model is/does. -->
|
||||
|
||||
|
||||
|
||||
## Model Details
|
||||
|
||||
### Model Description
|
||||
|
||||
<!-- Provide a longer summary of what this model is. -->
|
||||
|
||||
|
||||
|
||||
- **Developed by:** [More Information Needed]
|
||||
- **Funded by [optional]:** [More Information Needed]
|
||||
- **Shared by [optional]:** [More Information Needed]
|
||||
- **Model type:** [More Information Needed]
|
||||
- **Language(s) (NLP):** [More Information Needed]
|
||||
- **License:** [More Information Needed]
|
||||
- **Finetuned from model [optional]:** [More Information Needed]
|
||||
|
||||
### Model Sources [optional]
|
||||
|
||||
<!-- Provide the basic links for the model. -->
|
||||
|
||||
- **Repository:** [More Information Needed]
|
||||
- **Paper [optional]:** [More Information Needed]
|
||||
- **Demo [optional]:** [More Information Needed]
|
||||
|
||||
## Uses
|
||||
|
||||
<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
|
||||
|
||||
### Direct Use
|
||||
|
||||
<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
### Downstream Use [optional]
|
||||
|
||||
<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
### Out-of-Scope Use
|
||||
|
||||
<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Bias, Risks, and Limitations
|
||||
|
||||
<!-- This section is meant to convey both technical and sociotechnical limitations. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
### Recommendations
|
||||
|
||||
<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
|
||||
|
||||
Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
|
||||
|
||||
## How to Get Started with the Model
|
||||
|
||||
Use the code below to get started with the model.
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Training Details
|
||||
|
||||
### Training Data
|
||||
|
||||
<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
### Training Procedure
|
||||
|
||||
<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
|
||||
|
||||
#### Preprocessing [optional]
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
|
||||
#### Training Hyperparameters
|
||||
|
||||
- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
|
||||
|
||||
#### Speeds, Sizes, Times [optional]
|
||||
|
||||
<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Evaluation
|
||||
|
||||
<!-- This section describes the evaluation protocols and provides the results. -->
|
||||
|
||||
### Testing Data, Factors & Metrics
|
||||
|
||||
#### Testing Data
|
||||
|
||||
<!-- This should link to a Dataset Card if possible. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
#### Factors
|
||||
|
||||
<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
#### Metrics
|
||||
|
||||
<!-- These are the evaluation metrics being used, ideally with a description of why. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
### Results
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
#### Summary
|
||||
|
||||
|
||||
|
||||
## Model Examination [optional]
|
||||
|
||||
<!-- Relevant interpretability work for the model goes here -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Environmental Impact
|
||||
|
||||
<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
|
||||
|
||||
Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
|
||||
|
||||
- **Hardware Type:** [More Information Needed]
|
||||
- **Hours used:** [More Information Needed]
|
||||
- **Cloud Provider:** [More Information Needed]
|
||||
- **Compute Region:** [More Information Needed]
|
||||
- **Carbon Emitted:** [More Information Needed]
|
||||
|
||||
## Technical Specifications [optional]
|
||||
|
||||
### Model Architecture and Objective
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
### Compute Infrastructure
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
#### Hardware
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
#### Software
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Citation [optional]
|
||||
|
||||
<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
|
||||
|
||||
**BibTeX:**
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
**APA:**
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Glossary [optional]
|
||||
|
||||
<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## More Information [optional]
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Model Card Authors [optional]
|
||||
|
||||
[More Information Needed]
|
||||
|
||||
## Model Card Contact
|
||||
|
||||
[More Information Needed]
|
||||
### Framework versions
|
||||
|
||||
- PEFT 0.19.1
|
||||
+43
@@ -0,0 +1,43 @@
|
||||
{
|
||||
"alora_invocation_tokens": null,
|
||||
"alpha_pattern": {},
|
||||
"arrow_config": null,
|
||||
"auto_mapping": null,
|
||||
"base_model_name_or_path": "HuggingFaceTB/SmolLM2-135M-Instruct",
|
||||
"bias": "none",
|
||||
"corda_config": null,
|
||||
"ensure_weight_tying": false,
|
||||
"eva_config": null,
|
||||
"exclude_modules": null,
|
||||
"fan_in_fan_out": false,
|
||||
"inference_mode": true,
|
||||
"init_lora_weights": true,
|
||||
"layer_replication": null,
|
||||
"layers_pattern": null,
|
||||
"layers_to_transform": null,
|
||||
"loftq_config": {},
|
||||
"lora_alpha": 32,
|
||||
"lora_bias": false,
|
||||
"lora_dropout": 0.0,
|
||||
"lora_ga_config": null,
|
||||
"megatron_config": null,
|
||||
"megatron_core": "megatron.core",
|
||||
"modules_to_save": null,
|
||||
"peft_type": "LORA",
|
||||
"peft_version": "0.19.1",
|
||||
"qalora_group_size": 16,
|
||||
"r": 16,
|
||||
"rank_pattern": {},
|
||||
"revision": null,
|
||||
"target_modules": [
|
||||
"q_proj",
|
||||
"v_proj"
|
||||
],
|
||||
"target_parameters": null,
|
||||
"task_type": "CAUSAL_LM",
|
||||
"trainable_token_indices": null,
|
||||
"use_bdlora": null,
|
||||
"use_dora": false,
|
||||
"use_qalora": false,
|
||||
"use_rslora": false
|
||||
}
|
||||
+6
@@ -0,0 +1,6 @@
|
||||
{% for message in messages %}{% if loop.first and messages[0]['role'] != 'system' %}{{ '<|im_start|>system
|
||||
You are a helpful AI assistant named SmolLM, trained by Hugging Face<|im_end|>
|
||||
' }}{% endif %}{{'<|im_start|>' + message['role'] + '
|
||||
' + message['content'] + '<|im_end|>' + '
|
||||
'}}{% endfor %}{% if add_generation_prompt %}{{ '<|im_start|>assistant
|
||||
' }}{% endif %}
|
||||
+19
@@ -0,0 +1,19 @@
|
||||
{
|
||||
"add_prefix_space": false,
|
||||
"backend": "tokenizers",
|
||||
"bos_token": "<|im_start|>",
|
||||
"clean_up_tokenization_spaces": false,
|
||||
"eos_token": "<|im_end|>",
|
||||
"errors": "replace",
|
||||
"extra_special_tokens": [
|
||||
"<|im_start|>",
|
||||
"<|im_end|>"
|
||||
],
|
||||
"is_local": false,
|
||||
"local_files_only": false,
|
||||
"model_max_length": 8192,
|
||||
"pad_token": "<|im_end|>",
|
||||
"tokenizer_class": "GPT2Tokenizer",
|
||||
"unk_token": "<|endoftext|>",
|
||||
"vocab_size": 49152
|
||||
}
|
||||
+80
@@ -0,0 +1,80 @@
|
||||
{"id": "test-ar-000", "gold_label": "ar", "source_label": "ar", "prediction": null, "response": "إن الاستم", "correct": false, "input_tokens": 114, "output_tokens": 8, "latency_seconds": 0.06024326104670763}
|
||||
{"id": "test-ar-001", "gold_label": "ar", "source_label": "ar", "prediction": null, "response": "أن يكون", "correct": false, "input_tokens": 156, "output_tokens": 8, "latency_seconds": 0.06023324280977249}
|
||||
{"id": "test-ar-002", "gold_label": "ar", "source_label": "ar", "prediction": null, "response": "أن يكون", "correct": false, "input_tokens": 156, "output_tokens": 8, "latency_seconds": 0.05917225405573845}
|
||||
{"id": "test-ar-003", "gold_label": "ar", "source_label": "ar", "prediction": null, "response": "هذا هو �", "correct": false, "input_tokens": 91, "output_tokens": 8, "latency_seconds": 0.059281811118125916}
|
||||
{"id": "test-bg-000", "gold_label": "ot", "source_label": "bg", "prediction": null, "response": "Освен че", "correct": false, "input_tokens": 147, "output_tokens": 8, "latency_seconds": 0.059322684071958065}
|
||||
{"id": "test-bg-001", "gold_label": "ot", "source_label": "bg", "prediction": null, "response": "Общата би", "correct": false, "input_tokens": 69, "output_tokens": 8, "latency_seconds": 0.0591945294290781}
|
||||
{"id": "test-bg-002", "gold_label": "ot", "source_label": "bg", "prediction": null, "response": "Тънкост", "correct": false, "input_tokens": 118, "output_tokens": 8, "latency_seconds": 0.059487082064151764}
|
||||
{"id": "test-bg-003", "gold_label": "ot", "source_label": "bg", "prediction": null, "response": "Просматри", "correct": false, "input_tokens": 85, "output_tokens": 8, "latency_seconds": 0.05959406774491072}
|
||||
{"id": "test-de-000", "gold_label": "de", "source_label": "de", "prediction": null, "response": "Die Abdeckung (dünnes", "correct": false, "input_tokens": 174, "output_tokens": 8, "latency_seconds": 0.05979781597852707}
|
||||
{"id": "test-de-001", "gold_label": "de", "source_label": "de", "prediction": null, "response": "Besser etwas größ", "correct": false, "input_tokens": 43, "output_tokens": 8, "latency_seconds": 0.05983380135148764}
|
||||
{"id": "test-de-002", "gold_label": "de", "source_label": "de", "prediction": null, "response": "Schade, wie könn", "correct": false, "input_tokens": 60, "output_tokens": 8, "latency_seconds": 0.05958701949566603}
|
||||
{"id": "test-de-003", "gold_label": "de", "source_label": "de", "prediction": null, "response": "Dich und Robust, alles", "correct": false, "input_tokens": 43, "output_tokens": 8, "latency_seconds": 0.059297237545251846}
|
||||
{"id": "test-el-000", "gold_label": "el", "source_label": "el", "prediction": null, "response": "Το Κρίσ", "correct": false, "input_tokens": 165, "output_tokens": 8, "latency_seconds": 0.059339928440749645}
|
||||
{"id": "test-el-001", "gold_label": "el", "source_label": "el", "prediction": null, "response": "Πολιτεί", "correct": false, "input_tokens": 111, "output_tokens": 8, "latency_seconds": 0.05949285626411438}
|
||||
{"id": "test-el-002", "gold_label": "el", "source_label": "el", "prediction": null, "response": "Και είναι δ", "correct": false, "input_tokens": 92, "output_tokens": 8, "latency_seconds": 0.059603940695524216}
|
||||
{"id": "test-el-003", "gold_label": "el", "source_label": "el", "prediction": null, "response": "Και είναι δ", "correct": false, "input_tokens": 92, "output_tokens": 8, "latency_seconds": 0.05931456387042999}
|
||||
{"id": "test-en-000", "gold_label": "en", "source_label": "en", "prediction": null, "response": "I'm sorry for the confusion, but", "correct": false, "input_tokens": 50, "output_tokens": 8, "latency_seconds": 0.059724729508161545}
|
||||
{"id": "test-en-001", "gold_label": "en", "source_label": "en", "prediction": null, "response": "I'm sorry for the misunderstanding, but", "correct": false, "input_tokens": 69, "output_tokens": 8, "latency_seconds": 0.05924297962337732}
|
||||
{"id": "test-en-002", "gold_label": "en", "source_label": "en", "prediction": null, "response": "I'm sorry for the inconvenience, but", "correct": false, "input_tokens": 81, "output_tokens": 8, "latency_seconds": 0.059761520475149155}
|
||||
{"id": "test-en-003", "gold_label": "en", "source_label": "en", "prediction": null, "response": "I’m sorry for the inconvenience,", "correct": false, "input_tokens": 57, "output_tokens": 8, "latency_seconds": 0.05949122551828623}
|
||||
{"id": "test-es-000", "gold_label": "es", "source_label": "es", "prediction": null, "response": "La compré para mi mujer", "correct": false, "input_tokens": 149, "output_tokens": 8, "latency_seconds": 0.05953405052423477}
|
||||
{"id": "test-es-001", "gold_label": "es", "source_label": "es", "prediction": null, "response": "No se sujeta bien al", "correct": false, "input_tokens": 63, "output_tokens": 8, "latency_seconds": 0.05952043551951647}
|
||||
{"id": "test-es-002", "gold_label": "es", "source_label": "es", "prediction": null, "response": "Hola, pero no es exact", "correct": false, "input_tokens": 134, "output_tokens": 8, "latency_seconds": 0.0594380721449852}
|
||||
{"id": "test-es-003", "gold_label": "es", "source_label": "es", "prediction": null, "response": "No me ha llegado el ped", "correct": false, "input_tokens": 72, "output_tokens": 8, "latency_seconds": 0.059360750019550323}
|
||||
{"id": "test-fr-000", "gold_label": "fr", "source_label": "fr", "prediction": null, "response": "Au bout de 1 mois", "correct": false, "input_tokens": 71, "output_tokens": 8, "latency_seconds": 0.05943280924111605}
|
||||
{"id": "test-fr-001", "gold_label": "fr", "source_label": "fr", "prediction": null, "response": "Livraison vraiment d", "correct": false, "input_tokens": 57, "output_tokens": 8, "latency_seconds": 0.05940195545554161}
|
||||
{"id": "test-fr-002", "gold_label": "fr", "source_label": "fr", "prediction": null, "response": "Copie chinoise de l'", "correct": false, "input_tokens": 85, "output_tokens": 8, "latency_seconds": 0.05936688371002674}
|
||||
{"id": "test-fr-003", "gold_label": "fr", "source_label": "fr", "prediction": null, "response": "Je vous aimez de v", "correct": false, "input_tokens": 65, "output_tokens": 8, "latency_seconds": 0.059392811730504036}
|
||||
{"id": "test-hi-000", "gold_label": "hi", "source_label": "hi", "prediction": null, "response": "इसलिए", "correct": false, "input_tokens": 86, "output_tokens": 8, "latency_seconds": 0.059750827960669994}
|
||||
{"id": "test-hi-001", "gold_label": "hi", "source_label": "hi", "prediction": null, "response": "इसलिए", "correct": false, "input_tokens": 86, "output_tokens": 8, "latency_seconds": 0.0595023175701499}
|
||||
{"id": "test-hi-002", "gold_label": "hi", "source_label": "hi", "prediction": null, "response": "और, ज़", "correct": false, "input_tokens": 153, "output_tokens": 8, "latency_seconds": 0.05994911026209593}
|
||||
{"id": "test-hi-003", "gold_label": "hi", "source_label": "hi", "prediction": null, "response": "उह, म�", "correct": false, "input_tokens": 124, "output_tokens": 8, "latency_seconds": 0.059514570981264114}
|
||||
{"id": "test-it-000", "gold_label": "ot", "source_label": "it", "prediction": null, "response": "Una persona stelleva di c", "correct": false, "input_tokens": 44, "output_tokens": 8, "latency_seconds": 0.05949304159730673}
|
||||
{"id": "test-it-001", "gold_label": "ot", "source_label": "it", "prediction": null, "response": "Diversi bambini saltano", "correct": false, "input_tokens": 49, "output_tokens": 8, "latency_seconds": 0.05926634185016155}
|
||||
{"id": "test-it-002", "gold_label": "ot", "source_label": "it", "prediction": null, "response": "Il mese scorso è mig", "correct": false, "input_tokens": 96, "output_tokens": 8, "latency_seconds": 0.05920985620468855}
|
||||
{"id": "test-it-003", "gold_label": "ot", "source_label": "it", "prediction": null, "response": "A per diem, or a day", "correct": false, "input_tokens": 44, "output_tokens": 8, "latency_seconds": 0.059299188666045666}
|
||||
{"id": "test-ja-000", "gold_label": "ot", "source_label": "ja", "prediction": null, "response": "SSDへシス", "correct": false, "input_tokens": 193, "output_tokens": 8, "latency_seconds": 0.05936930328607559}
|
||||
{"id": "test-ja-001", "gold_label": "ot", "source_label": "ja", "prediction": null, "response": "このこと�", "correct": false, "input_tokens": 100, "output_tokens": 8, "latency_seconds": 0.05920923221856356}
|
||||
{"id": "test-ja-002", "gold_label": "ot", "source_label": "ja", "prediction": null, "response": "もう、見る", "correct": false, "input_tokens": 121, "output_tokens": 8, "latency_seconds": 0.05940388608723879}
|
||||
{"id": "test-ja-003", "gold_label": "ot", "source_label": "ja", "prediction": null, "response": "なんだか", "correct": false, "input_tokens": 134, "output_tokens": 8, "latency_seconds": 0.059428741224110126}
|
||||
{"id": "test-nl-000", "gold_label": "ot", "source_label": "nl", "prediction": null, "response": "Het vlees, gev", "correct": false, "input_tokens": 76, "output_tokens": 8, "latency_seconds": 0.05941706243902445}
|
||||
{"id": "test-nl-001", "gold_label": "ot", "source_label": "nl", "prediction": null, "response": "Hoe zijn de kunst", "correct": false, "input_tokens": 46, "output_tokens": 8, "latency_seconds": 0.05952310189604759}
|
||||
{"id": "test-nl-002", "gold_label": "ot", "source_label": "nl", "prediction": null, "response": "In the water.", "correct": false, "input_tokens": 44, "output_tokens": 5, "latency_seconds": 0.03761201910674572}
|
||||
{"id": "test-nl-003", "gold_label": "ot", "source_label": "nl", "prediction": null, "response": "De technologisch geplaat", "correct": false, "input_tokens": 81, "output_tokens": 8, "latency_seconds": 0.05957785062491894}
|
||||
{"id": "test-pl-000", "gold_label": "ot", "source_label": "pl", "prediction": null, "response": "Opalona dziewczyna", "correct": false, "input_tokens": 64, "output_tokens": 8, "latency_seconds": 0.059507221914827824}
|
||||
{"id": "test-pl-001", "gold_label": "ot", "source_label": "pl", "prediction": null, "response": "I'm sorry for any misunderstanding, but", "correct": false, "input_tokens": 60, "output_tokens": 8, "latency_seconds": 0.0593439182266593}
|
||||
{"id": "test-pl-002", "gold_label": "ot", "source_label": "pl", "prediction": null, "response": "I'm sorry for the misunderstanding, but", "correct": false, "input_tokens": 71, "output_tokens": 8, "latency_seconds": 0.05937239807099104}
|
||||
{"id": "test-pl-003", "gold_label": "ot", "source_label": "pl", "prediction": null, "response": "Średnia dla prz", "correct": false, "input_tokens": 90, "output_tokens": 8, "latency_seconds": 0.059316896833479404}
|
||||
{"id": "test-pt-000", "gold_label": "ot", "source_label": "pt", "prediction": null, "response": "A Kollar-Kotelly ag", "correct": false, "input_tokens": 62, "output_tokens": 8, "latency_seconds": 0.059342941269278526}
|
||||
{"id": "test-pt-001", "gold_label": "ot", "source_label": "pt", "prediction": null, "response": "A jovem mulher está", "correct": false, "input_tokens": 51, "output_tokens": 8, "latency_seconds": 0.05939680803567171}
|
||||
{"id": "test-pt-002", "gold_label": "ot", "source_label": "pt", "prediction": null, "response": "Um grupo de homens f", "correct": false, "input_tokens": 46, "output_tokens": 8, "latency_seconds": 0.059792510233819485}
|
||||
{"id": "test-pt-003", "gold_label": "ot", "source_label": "pt", "prediction": null, "response": "Uma criança está", "correct": false, "input_tokens": 62, "output_tokens": 8, "latency_seconds": 0.05932790972292423}
|
||||
{"id": "test-ru-000", "gold_label": "ru", "source_label": "ru", "prediction": null, "response": "Моим детя", "correct": false, "input_tokens": 94, "output_tokens": 8, "latency_seconds": 0.05943390540778637}
|
||||
{"id": "test-ru-001", "gold_label": "ru", "source_label": "ru", "prediction": null, "response": "Но я торо�", "correct": false, "input_tokens": 55, "output_tokens": 8, "latency_seconds": 0.059396409429609776}
|
||||
{"id": "test-ru-002", "gold_label": "ru", "source_label": "ru", "prediction": null, "response": "Но я торо�", "correct": false, "input_tokens": 55, "output_tokens": 8, "latency_seconds": 0.05932094994932413}
|
||||
{"id": "test-ru-003", "gold_label": "ru", "source_label": "ru", "prediction": null, "response": "Через каж", "correct": false, "input_tokens": 104, "output_tokens": 8, "latency_seconds": 0.05943276919424534}
|
||||
{"id": "test-sw-000", "gold_label": "ot", "source_label": "sw", "prediction": null, "response": "Baada ya kukataa k", "correct": false, "input_tokens": 101, "output_tokens": 8, "latency_seconds": 0.05936537031084299}
|
||||
{"id": "test-sw-001", "gold_label": "ot", "source_label": "sw", "prediction": null, "response": "Kwa hiyo, watu", "correct": false, "input_tokens": 97, "output_tokens": 8, "latency_seconds": 0.05964792147278786}
|
||||
{"id": "test-sw-002", "gold_label": "ot", "source_label": "sw", "prediction": null, "response": "Ni nafasi yetu pekee", "correct": false, "input_tokens": 80, "output_tokens": 8, "latency_seconds": 0.05959350895136595}
|
||||
{"id": "test-sw-003", "gold_label": "ot", "source_label": "sw", "prediction": null, "response": "Kwa upande mwingine,", "correct": false, "input_tokens": 73, "output_tokens": 8, "latency_seconds": 0.059357658959925175}
|
||||
{"id": "test-th-000", "gold_label": "ot", "source_label": "th", "prediction": null, "response": "ชายหา", "correct": false, "input_tokens": 259, "output_tokens": 8, "latency_seconds": 0.06841117981821299}
|
||||
{"id": "test-th-001", "gold_label": "ot", "source_label": "th", "prediction": null, "response": "อย่า�", "correct": false, "input_tokens": 257, "output_tokens": 8, "latency_seconds": 0.060420410707592964}
|
||||
{"id": "test-th-002", "gold_label": "ot", "source_label": "th", "prediction": null, "response": "บ่อย", "correct": false, "input_tokens": 108, "output_tokens": 8, "latency_seconds": 0.0594320222735405}
|
||||
{"id": "test-th-003", "gold_label": "ot", "source_label": "th", "prediction": null, "response": "Farid Hilali is a renowned actor", "correct": false, "input_tokens": 122, "output_tokens": 8, "latency_seconds": 0.059269363060593605}
|
||||
{"id": "test-tr-000", "gold_label": "tr", "source_label": "tr", "prediction": null, "response": "Kayıt olarak, u", "correct": false, "input_tokens": 92, "output_tokens": 8, "latency_seconds": 0.05965294037014246}
|
||||
{"id": "test-tr-001", "gold_label": "tr", "source_label": "tr", "prediction": null, "response": "Kayıt olarak, u", "correct": false, "input_tokens": 92, "output_tokens": 8, "latency_seconds": 0.05943188723176718}
|
||||
{"id": "test-tr-002", "gold_label": "tr", "source_label": "tr", "prediction": null, "response": "Konuşul, dedi", "correct": false, "input_tokens": 71, "output_tokens": 8, "latency_seconds": 0.05948237422853708}
|
||||
{"id": "test-tr-003", "gold_label": "tr", "source_label": "tr", "prediction": null, "response": "Savaştan sonra yap", "correct": false, "input_tokens": 120, "output_tokens": 8, "latency_seconds": 0.05945493374019861}
|
||||
{"id": "test-ur-000", "gold_label": "ur", "source_label": "ur", "prediction": null, "response": "شروع دو", "correct": false, "input_tokens": 129, "output_tokens": 8, "latency_seconds": 0.059702896513044834}
|
||||
{"id": "test-ur-001", "gold_label": "ur", "source_label": "ur", "prediction": null, "response": "شرایط �", "correct": false, "input_tokens": 117, "output_tokens": 8, "latency_seconds": 0.05930099170655012}
|
||||
{"id": "test-ur-002", "gold_label": "ur", "source_label": "ur", "prediction": null, "response": "انٹیلی ج", "correct": false, "input_tokens": 166, "output_tokens": 8, "latency_seconds": 0.05926219932734966}
|
||||
{"id": "test-ur-003", "gold_label": "ur", "source_label": "ur", "prediction": null, "response": "دیکھو �", "correct": false, "input_tokens": 64, "output_tokens": 8, "latency_seconds": 0.05909104458987713}
|
||||
{"id": "test-vi-000", "gold_label": "vi", "source_label": "vi", "prediction": null, "response": "Người c", "correct": false, "input_tokens": 123, "output_tokens": 8, "latency_seconds": 0.059413446113467216}
|
||||
{"id": "test-vi-001", "gold_label": "vi", "source_label": "vi", "prediction": null, "response": "Đây là 5", "correct": false, "input_tokens": 176, "output_tokens": 8, "latency_seconds": 0.0593422232195735}
|
||||
{"id": "test-vi-002", "gold_label": "vi", "source_label": "vi", "prediction": null, "response": "Nghiệp v�", "correct": false, "input_tokens": 66, "output_tokens": 8, "latency_seconds": 0.05937460344284773}
|
||||
{"id": "test-vi-003", "gold_label": "vi", "source_label": "vi", "prediction": null, "response": "Bạn đã �", "correct": false, "input_tokens": 103, "output_tokens": 8, "latency_seconds": 0.05925013776868582}
|
||||
{"id": "test-zh-000", "gold_label": "zh", "source_label": "zh", "prediction": null, "response": "根本不是那", "correct": false, "input_tokens": 85, "output_tokens": 8, "latency_seconds": 0.059398721903562546}
|
||||
{"id": "test-zh-001", "gold_label": "zh", "source_label": "zh", "prediction": null, "response": "邮寄地�", "correct": false, "input_tokens": 109, "output_tokens": 8, "latency_seconds": 0.059589968994259834}
|
||||
{"id": "test-zh-002", "gold_label": "zh", "source_label": "zh", "prediction": null, "response": "面料很好", "correct": false, "input_tokens": 99, "output_tokens": 8, "latency_seconds": 0.05934322252869606}
|
||||
{"id": "test-zh-003", "gold_label": "zh", "source_label": "zh", "prediction": null, "response": "我订购的是", "correct": false, "input_tokens": 69, "output_tokens": 8, "latency_seconds": 0.05917076487094164}
|
||||
+14
@@ -0,0 +1,14 @@
|
||||
{
|
||||
"arm": "baseline",
|
||||
"rows": 80,
|
||||
"correct": 0,
|
||||
"accuracy": 0.0,
|
||||
"parse_rate": 0.0,
|
||||
"input_tokens": 7738,
|
||||
"output_tokens": 637,
|
||||
"latency_seconds_total": 4.744735201820731,
|
||||
"latency_seconds_mean": 0.05930919002275914,
|
||||
"latency_seconds_median": 0.05941525427624583,
|
||||
"provider_charge_usd": 0.0,
|
||||
"provider_charge_note": "Local inference made no provider API calls; electricity/hardware amortization is excluded."
|
||||
}
|
||||
+160
@@ -0,0 +1,160 @@
|
||||
{"id": "train-ar-000", "messages": [{"role": "user", "content": "اسفل على اليمين مطعم ماكسيم , الذي بدا ك a صالون ايس كريم وهو الان a نصب اكثر الموقر من مادلين ."}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876c0b8d682056dbef7b", "gold_label": "ar"}
|
||||
{"id": "train-ar-001", "messages": [{"role": "user", "content": "هل تحب الافلام المعلقة او هل تحب فقط اكشن او"}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876c3d093ac09861553b", "gold_label": "ar"}
|
||||
{"id": "train-ar-002", "messages": [{"role": "user", "content": "اعاد الموقع فورا موقع برادلي الرسمي , موقع لاخبار عن السباق الرئاسي لعام 2000 , معلومات عن البطاقة الطبية , والكتب عن برادلي ."}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876c7d03af917e2edc1a", "gold_label": "ar"}
|
||||
{"id": "train-ar-003", "messages": [{"role": "user", "content": "هل انت مشترك في اي هندسة رسم الاشياء التي"}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876cf4fe3c88d68d68fd", "gold_label": "ar"}
|
||||
{"id": "train-ar-004", "messages": [{"role": "user", "content": "00-187-الاثار المترتبة على الضمان الاجتماعي بالنسبة للمعاشات التقاعدية الخاصة ( جاو / hehs-00-187 , 14 ايلول / سبتمبر 2000 ) ."}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876c571e6ffa0b9cab99", "gold_label": "ar"}
|
||||
{"id": "train-ar-005", "messages": [{"role": "user", "content": "ويشمل موقع anao وصلات بمنشوراتها المختلفة , بما في ذلك تقارير مراجعة الحسابات و يهدي الممارسات الافضل ."}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876cb9c19ffa82a80140", "gold_label": "ar"}
|
||||
{"id": "train-ar-006", "messages": [{"role": "user", "content": "شاليت يقول انه عندما تمشي في الشارع يمكنك ان تخبر العذارى بسبب وهجهم الطازج والمتوازن ."}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876cde3406f5dc3ef752", "gold_label": "ar"}
|
||||
{"id": "train-ar-007", "messages": [{"role": "user", "content": "- ويكاد يكون تصنيع حافزا حافزا مكرسا تماما لتطبيقات توليد الطاقة ."}, {"role": "assistant", "content": "ar"}], "teacher_response_id": "chatcmpl-6a6a876cde3406f5dc3ef753", "gold_label": "ar"}
|
||||
{"id": "train-bg-000", "messages": [{"role": "user", "content": "това е на описание на г-н браун !"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876cbc7cc35e30f1df0a", "gold_label": "ot"}
|
||||
{"id": "train-bg-001", "messages": [{"role": "user", "content": "за финансовата 2002 година vba преразгледа плановете си за ефективност на висшите ръководители в регионалните служби за подобряване на индивидуалната отчетност за елементите на изпълнение чрез свързване на целите на организацията за ефективност и действителните резултати с значими и измерими елементи на ефективност ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876ca5c09dd52de3930e", "gold_label": "ot"}
|
||||
{"id": "train-bg-002", "messages": [{"role": "user", "content": "журналистът е измамник , предназначен само да кара трафика до порно сайтовете на собственика ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876cccc0dbe1b98e03c1", "gold_label": "ot"}
|
||||
{"id": "train-bg-003", "messages": [{"role": "user", "content": "толкова добре , колкото хората казват , че е ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876cb9c19ffa82a8013f", "gold_label": "ot"}
|
||||
{"id": "train-bg-004", "messages": [{"role": "user", "content": "о , имаш ли такова време в северна каролина ?"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876eb9c19ffa82a80142", "gold_label": "ot"}
|
||||
{"id": "train-bg-005", "messages": [{"role": "user", "content": "както и да е , закуската е била принуден в тях сред жените преди известно време , така че няма за какво да се тревожим ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876fd1bc7b6a3fb5f715", "gold_label": "ot"}
|
||||
{"id": "train-bg-006", "messages": [{"role": "user", "content": "тийнейджърите биха искали да се присъединят към на диско орди в хараджуку , близо до парка yoyogi ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876f419b7910c2fd95d3", "gold_label": "ot"}
|
||||
{"id": "train-bg-007", "messages": [{"role": "user", "content": "циници стенеше , че хитчинс се е цел в клинтън , но вместо това е застрелял блументал ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a876f7d03af917e2edc1d", "gold_label": "ot"}
|
||||
{"id": "train-de-000", "messages": [{"role": "user", "content": "Sieht gut aus und hat super an meinen skoda Schlüssel gepasst"}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a87708ff452ff67d52a45", "gold_label": "de"}
|
||||
{"id": "train-de-001", "messages": [{"role": "user", "content": "Kann mich obenstehenden Rezensionen nur anschließen, das Headset ist im Grunde genommen sein Geld wert, würde es nicht nach einem Jahr mit Wackelkontakt den Geist aufgeben."}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a87705a513bb219e7134f", "gold_label": "de"}
|
||||
{"id": "train-de-002", "messages": [{"role": "user", "content": "Eine Flasche ist leider mehr oder weniger ausgelaufen. Nichtsdestotrotz tut die Kochsalzlösung was sie soll. Punktabzug gibt es nur wegen dem auslaufen."}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a87703d093ac09861553e", "gold_label": "de"}
|
||||
{"id": "train-de-003", "messages": [{"role": "user", "content": "Habe die Tasche als Geschenk gekauft - Lieferung funktionierte spitze und die Tasche sieht wirklich toll aus (Verarbeitung, Stoff usw.! Daumen hoch!"}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a8770bcb8b1ca0dc259ad", "gold_label": "de"}
|
||||
{"id": "train-de-004", "messages": [{"role": "user", "content": "Gut finde ich die Optik und die Ersatzbürsten. Nicht ganz durchdacht finde ich den Behälter, wo die Klobürste rein kommt. Nimmt man diese raus und steckt sie wieder ein, reibt der Behälter über den Fliesenboden (den man üblicherweise in der Toilette hat) was was Geräuche macht und einen schlechten Qualitätseindruck hinterlässt. Sinnvoll wäre, wenn auf der Unterseite noch eine Gummifläche geklebt würde, so dass Metall nicht auf Fliesen reibt."}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a8770a5c09dd52de39312", "gold_label": "de"}
|
||||
{"id": "train-de-005", "messages": [{"role": "user", "content": "Der Hut kam sehr spät und fällt ziemlich klein aus. Er sieht aber gut aus und kann praktisch zusammen geklappt werden. Für den Preis - alles in Ordnung."}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a8771adf8be004bbe4b02", "gold_label": "de"}
|
||||
{"id": "train-de-006", "messages": [{"role": "user", "content": "Das Kleidchen ist wirklich schön. Allerdings wirklich nur was für den Strand. Ich trage normalerweise eine S. Hab mir eine M bestellt, damit es etwas lockerer sitzt, aber ich würde sagen, dass sogar eine L angebracht wäre."}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a8771f533e66c8281caac", "gold_label": "de"}
|
||||
{"id": "train-de-007", "messages": [{"role": "user", "content": "Ich hatte diese Modell vor fünf Jahren gekauft. War damals super Qualität und hat auch lange gehalten. Jetzt das gleiche Modell von HAMA hier nachgekauft und große Enttäuschung. Kabel viel dünner als vor fünf Jahren. Verarbeitung viel schlechter. Nur günstige Materialien verwendet. Habe das Headset nun schon zum zweiten Mal umgetauscht und eine neues erhalten. Im täglichen Gebrauch hält es etwa zwei Wochen, dann schauen überall die Drähte raus und die Plastikteile lösen sich. Leider nicht mehr zu empfehlen !!!"}, {"role": "assistant", "content": "de"}], "teacher_response_id": "chatcmpl-6a6a87713f3c9b5f06007edc", "gold_label": "de"}
|
||||
{"id": "train-el-000", "messages": [{"role": "user", "content": "ΠΊΝΑΚΑΣ 1 : σύγκριση των ηγετικών πρακτικών και των ομοσπονδιακών πρακτικών διαχείρισης των cio"}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a8772a5c09dd52de39314", "gold_label": "el"}
|
||||
{"id": "train-el-001", "messages": [{"role": "user", "content": "Η ευελιξία του προϋπολογισμού μειώνεται δραστικά , έτσι ώστε μέχρι το 2050 , το καθαρό ενδιαφέρον για το χρέος θα απορροφήσει περίπου το ήμισυ όλων των ομοσπονδιακών εσόδων ."}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a87733165d2c706d08be3", "gold_label": "el"}
|
||||
{"id": "train-el-002", "messages": [{"role": "user", "content": "Για όλο το ιστορικό μεγαλείο της , η πιστή γερουσία τώρα είναι το ισοδύναμο ενός Δημοτικού Συμβουλίου , η πολιτική του ικανότητα αφιερωμένη στις προμήθειες νερού , τις γραμμές αποχέτευσης και την ίδρυση παιδότοπους ."}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a8773bfb3431a6ebc7f7d", "gold_label": "el"}
|
||||
{"id": "train-el-003", "messages": [{"role": "user", "content": "Θα πρέπει να πληρούνται οι απαιτήσεις για την κατάρτιση , την τεκμηρίωση και τη συντήρηση ."}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a87743d093ac098615542", "gold_label": "el"}
|
||||
{"id": "train-el-004", "messages": [{"role": "user", "content": "Προστατευμένη από τους κρύο , υγρασία , βορειοδυτικά ανέμους από τα βουνά των vosges , οι αμπελώνες της αλσατία απολαμβάνουν ένα ιδανικό μικροκλίμα για την παραγωγή λευκών κρασιών που έχουν την εμπιστοσύνη τους στα πιο διάσημα κρασιά της βουργουνδίας και του μπορντώ ."}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a877415c0ba78d0881f28", "gold_label": "el"}
|
||||
{"id": "train-el-005", "messages": [{"role": "user", "content": "Και κάναμε πολλή κηπουρική εκεί έξω και κυρίως σε μεγάλωσε κρεβάτια για να κρατήσουμε το χώμα ωραίο ."}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a87754d654f2ee1729b34", "gold_label": "el"}
|
||||
{"id": "train-el-006", "messages": [{"role": "user", "content": "Αχ μου αρέσει η γυναίκα μου δεν μπορεί να καταλάβει ότι θα είναι εκατό βαθμούς έξω θα είμαι εκεί έξω αλλά κάνει πολύ ζέστη για να δουλεύω στην αυλή είμαι κάτω από τα δέντρα περνάω καλά και σταματάω να πιω νερό οπότε τι είναι το πρόβλημα ."}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a87740058eb9f85463804", "gold_label": "el"}
|
||||
{"id": "train-el-007", "messages": [{"role": "user", "content": "Απέναντι από το ο . Στο Audoen , δύο ενδιαφέροντες δρόμοι τρέχουν από την οδό Χάι ."}, {"role": "assistant", "content": "el"}], "teacher_response_id": "chatcmpl-6a6a8774b9c19ffa82a80147", "gold_label": "el"}
|
||||
{"id": "train-en-000", "messages": [{"role": "user", "content": "This was ordered as a pack of 2 but I only received 1. Emailed the seller but never received a reply. Feeling jipped."}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a8775220baf5a77ff2236", "gold_label": "en"}
|
||||
{"id": "train-en-001", "messages": [{"role": "user", "content": "Product received in damaged condition. Realized extent of damage after assembly almost completed since came in components. Unable to contact for product return."}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a8774157a4e90b05bc576", "gold_label": "en"}
|
||||
{"id": "train-en-002", "messages": [{"role": "user", "content": "Much cheaper quality than expected but item did arrive very fast and not damaged."}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a87755e7b32b0b9798f24", "gold_label": "en"}
|
||||
{"id": "train-en-003", "messages": [{"role": "user", "content": "The book started out well and then just went on and on and on....... and never quite got to the point until I just quit reading it. I got to about 60% of the way and simply could not continue. I see where others have simply skipped ahead full chapters to get to the last 15% of the book. By the time I gave up my interest had simply disappeared. Too bad, could have been written in a much more concise and interesting format."}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a8775bcb8b1ca0dc259b5", "gold_label": "en"}
|
||||
{"id": "train-en-004", "messages": [{"role": "user", "content": "Looked like it was going to be another homage to 80's horror in the vein of Ti West but it just doesn't know what it wants to be and wraps everything up in a really sloppy way. Setup, location and atmosphere are all there they just didn't know what to do with all of that after they jumped into the actual story. Kind of a disappointing miss since that is my only real complaint. All of the pieces were there for a great throwback horror movie and it was all squandered"}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a87750b8d682056dbef86", "gold_label": "en"}
|
||||
{"id": "train-en-005", "messages": [{"role": "user", "content": "Perfect for adding light layers of hydration especially in the colder months. This is usually my first layer of hydration before a moisturizer and it works great with no pilling or balling. Hydration lasts all day!"}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a877644ae93bbb7c864f2", "gold_label": "en"}
|
||||
{"id": "train-en-006", "messages": [{"role": "user", "content": "Very big and gaudy looking. Will not wear them."}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a87766a87287a70e858ce", "gold_label": "en"}
|
||||
{"id": "train-en-007", "messages": [{"role": "user", "content": "Product description says gown but its an all in one pants and way top big for a newborn. The headband stitching is coming undone, very disappointed."}, {"role": "assistant", "content": "en"}], "teacher_response_id": "chatcmpl-6a6a87760b8d682056dbef87", "gold_label": "en"}
|
||||
{"id": "train-es-000", "messages": [{"role": "user", "content": "Por lo que vale y con premium te sacas la compra del mes. Trae funda y dos protectores por un gran precio."}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a8777577a2132f2c86b03", "gold_label": "es"}
|
||||
{"id": "train-es-001", "messages": [{"role": "user", "content": "Son finas de algodón 100%, ideales para el veranito, yo las he comprado para regalar a una amiga que ha dado a luz hace un par de meses."}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a8778e3dfcfab6fd10197", "gold_label": "es"}
|
||||
{"id": "train-es-002", "messages": [{"role": "user", "content": "Buenísimos. Pintan genial. Los he probado sobre cristal y tienen una solidez que permite que se vean perfectamente. Se mantienen durante mucho tiempo pero a la vez de borran con facilidad. Los recomiendo 100%. De hecho será la marca que compre a partir de ahora."}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a87777d03af917e2edc21", "gold_label": "es"}
|
||||
{"id": "train-es-003", "messages": [{"role": "user", "content": "Me ha gustado mucho tanto la calidad de los materiales como los acabados que tiene, el toque de la madera es muy chulo. Tiene las clásicas tres velocidades aunque como ya sabéis se usa en el 90% de las veces la misma. Se puede colocar unos altavoces por la parte de atrás, la verdad que suena bastante más alto de lo que me pensaba, aún así he conectado los dos altavoces que tengo repartidos por el comedor y el resultado es espectacular. Unos recuerdos del sonido clásico que me ha gustado mucho recordar, estoy muy contento con el resultado. A destacar la superficie de la maleta, estéticamente es muy bonita, incluso simplemente para tener colocado encima de algún mueble, cumple perfectamente los dos cometidos, la estética y la funcional. Por supuesto cuenta con entrada USB y con función bluetooth, perfecto porque puedes reproducir también la música del teléfono móvil."}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a8778dac2ed2f421575cc", "gold_label": "es"}
|
||||
{"id": "train-es-004", "messages": [{"role": "user", "content": "Muy robusto, era para niño de 8 años pero como es regulable le servirá para mucho"}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a8778db7d851280a77052", "gold_label": "es"}
|
||||
{"id": "train-es-005", "messages": [{"role": "user", "content": "Me gusta brillo de labios, es muy poca cantidad pero esta bien."}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a8778aff9f7f7f84cf43b", "gold_label": "es"}
|
||||
{"id": "train-es-006", "messages": [{"role": "user", "content": "No es compatible para el nuevo XS, todas las fundas son del X y las venden como XS y no saben que no es igual, solo hay que ver el altavoz de al rededor de la cámara"}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a87784d654f2ee1729b38", "gold_label": "es"}
|
||||
{"id": "train-es-007", "messages": [{"role": "user", "content": "Esta bien pero pense que incluía las bombillas..."}, {"role": "assistant", "content": "es"}], "teacher_response_id": "chatcmpl-6a6a8779124dbf4a23c9891a", "gold_label": "es"}
|
||||
{"id": "train-fr-000", "messages": [{"role": "user", "content": "Bien reçu bien emballer fonctionne correctement, j’avais peur du bruit par rapport au commentaires mais le bruit des ventilateurs est correct"}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a87790a128ec1a6782ce7", "gold_label": "fr"}
|
||||
{"id": "train-fr-001", "messages": [{"role": "user", "content": "Ce produit est tout simplement génial. il agit très vite. l'odeur disparait quasiment au moment de la pulvérisation, il n'y a pas pas besoin d'en mettre beaucoup le flacon dure longtemps. Ensuite le nettoyage est très facile et il y a même comme une petite odeur de frais. Le chat ne reviens pas faire au même endroit après avoir nettoyé avec ce produit. N'hésitez pas."}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a877aa22fab7265b5be39", "gold_label": "fr"}
|
||||
{"id": "train-fr-002", "messages": [{"role": "user", "content": "Beau mais pour coller c'est pas très efficace. J'en ai essayé 2 ils n'ont pas tenu 1 semaine. Mais j'en ai fixé un avec une vis. Donc on peut l'adapter."}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a877912e92b55023aa874", "gold_label": "fr"}
|
||||
{"id": "train-fr-003", "messages": [{"role": "user", "content": "L' idée m' avait séduit mais l' objet est décevant. Les chiffres sont de simples autocollants , la rotation des pièces n' est pas fluide. Je n' aurais pas acheté si j' avais pu le toucher."}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a87793165d2c706d08bee", "gold_label": "fr"}
|
||||
{"id": "train-fr-004", "messages": [{"role": "user", "content": "Torche puissante, idéale pour un vélo, le système de fixation semble fiable. En revanche, je regrette la batterie fournie et nécessaire qui ne correspond à aucune classe de pile classique."}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a877b98f2260e94fe9ad9", "gold_label": "fr"}
|
||||
{"id": "train-fr-005", "messages": [{"role": "user", "content": "LIVRAISON RAPIDE ET SERIEUSE"}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a877bd657b69fbc2d80a5", "gold_label": "fr"}
|
||||
{"id": "train-fr-006", "messages": [{"role": "user", "content": "Je viens de m'apercevoir que ce produit vendu comme neuf semble être reconditionné puisque une photo est restée dans l'appareil (personnes et lieu que je ne connais pas du tout) . La date de la photo est antérieure à ma date d'achat."}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a877b15c0ba78d0881f2d", "gold_label": "fr"}
|
||||
{"id": "train-fr-007", "messages": [{"role": "user", "content": "Simple jouet avec lequel on s'amuse deux minutes avec , la mode est finis"}, {"role": "assistant", "content": "fr"}], "teacher_response_id": "chatcmpl-6a6a877b419b7910c2fd95e0", "gold_label": "fr"}
|
||||
{"id": "train-hi-000", "messages": [{"role": "user", "content": "हम प ् रत ् येक राज ् य के लिए कर कोड और प ् रक ् रियाओं को प ् राप ् त कर सकते हैं , इन ् हें जांच करें , चुनिंदा अधिकारियों का साक ् षात ् कार करें और कुछ विश ् वसनीय पैटर ् न उत ् पन ् न करें ."}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877ba5c09dd52de3931d", "gold_label": "hi"}
|
||||
{"id": "train-hi-001", "messages": [{"role": "user", "content": "और उम हाँ उह हाँ मुझे लगता है कि मैं आमतौर पर मैं हूँ उम भारी सॉस और उम"}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877ce5cd1f53d10ec52e", "gold_label": "hi"}
|
||||
{"id": "train-hi-002", "messages": [{"role": "user", "content": "जून में , जो कि जून में अपेक ् षित है , वह व ् हाइट हाउस वकील के जनादेश के न ् यायालय की व ् याख ् या को चालू करेगा . इस निर ् णय का कोई कानूनी उदाहरण नहीं है ."}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877cbe759e5d2a00e61e", "gold_label": "hi"}
|
||||
{"id": "train-hi-003", "messages": [{"role": "user", "content": "Anao की साइट में उनके विभिन ् न प ् रकाशनों के लिंक शामिल हैं , जिसमें अपनी लेखापरीक ् षा रिपोर ् ट और बेहतर अभ ् यास गाइड शामिल हैं ."}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877cde3406f5dc3ef76d", "gold_label": "hi"}
|
||||
{"id": "train-hi-004", "messages": [{"role": "user", "content": "11 वीं राजवंश की अवधि से पीटर के प ् राचीन मिस ् री धर ् म , सी . 2134 ईसा पूर ् व"}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877e628dc39d96c21c28", "gold_label": "hi"}
|
||||
{"id": "train-hi-005", "messages": [{"role": "user", "content": "हाँ हाँ किसी ने कहा कि उम कि इस तरह के कुछ प ् रकार के शुरू में कुछ प ् रकार की है और यह कि जब कार ठंडा हो जाता है , तो धातु को छोटा बनाने के लिए फैलता है और तब तक यह नहीं है कि कार को नीचे ठंडा होने तक नहीं है धातु आप जानते हैं कि इन दो टुकड ़ े किसी और को छू नहीं रहे हैं और आप और आप जानते हैं कि आप जानते हैं तो यह शुरू हो जाएगा क ् योंकि आपके पास छोटा नहीं है"}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877d4cb5f60e0c84468e", "gold_label": "hi"}
|
||||
{"id": "train-hi-006", "messages": [{"role": "user", "content": "लेखा विषयों के अतिरिक ् त , निर ् देशिका सूची में कुछ ऐसी एजेंसियों या प ् रोग ् राम जो उदाहरण में प ् रयोग किया गया है या उसके मानकों के भीतर अद ् वितीय प ् रावधान है ."}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877dbcc40607e9f8ab29", "gold_label": "hi"}
|
||||
{"id": "train-hi-007", "messages": [{"role": "user", "content": "वे चीज ़ ें जो केवल आप को प ् रभावित कर रहे हैं , उम कि हम लोगों को कैसे दंडित करते हैं और क ् यों चीजें हैं जैसे कि वे रास ् ते हैं"}, {"role": "assistant", "content": "hi"}], "teacher_response_id": "chatcmpl-6a6a877e628dc39d96c21c2a", "gold_label": "hi"}
|
||||
{"id": "train-it-000", "messages": [{"role": "user", "content": "Gli aerei da guerra russi colpiscono in Siria"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a877e4cff86af330ad142", "gold_label": "ot"}
|
||||
{"id": "train-it-001", "messages": [{"role": "user", "content": "L'uomo sta portando una cassetta degli attrezzi sul marciapiede."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a877eccc0dbe1b98e03e2", "gold_label": "ot"}
|
||||
{"id": "train-it-002", "messages": [{"role": "user", "content": "Israele trattiene 37 palestinesi mentre prosegue l'operazione di arresto"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a877f9904a4c085eca381", "gold_label": "ot"}
|
||||
{"id": "train-it-003", "messages": [{"role": "user", "content": "Uomo ucciso in un raid del terrore francese"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a877f80beffa1ad578f66", "gold_label": "ot"}
|
||||
{"id": "train-it-004", "messages": [{"role": "user", "content": "La gente va e pagaia su una zattera."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87805657fe29c59a3a5c", "gold_label": "ot"}
|
||||
{"id": "train-it-005", "messages": [{"role": "user", "content": "Mi oppongo alla pena di morte."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87819eab2b3298a1c572", "gold_label": "ot"}
|
||||
{"id": "train-it-006", "messages": [{"role": "user", "content": "La donna sta condendo l'olio."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8780567e9c2404b6c4c6", "gold_label": "ot"}
|
||||
{"id": "train-it-007", "messages": [{"role": "user", "content": "In una dichiarazione via e-mail al Knoxville News Sentinel, Shumaker ha detto: \"Non prendo in considerazione le dimissioni."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87813165d2c706d08c01", "gold_label": "ot"}
|
||||
{"id": "train-ja-000", "messages": [{"role": "user", "content": "味はとても良かったのですが、もうちょっと手頃な価格だと良かったのですが・・・"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87815a513bb219e71373", "gold_label": "ot"}
|
||||
{"id": "train-ja-001", "messages": [{"role": "user", "content": "ジュースに入れて使用しています。まだ加熱料理にしようしていませんが、機会があれば使用してみたいと思います。"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878233bb1483514480b9", "gold_label": "ot"}
|
||||
{"id": "train-ja-002", "messages": [{"role": "user", "content": "７キロ用の洗濯機にはパツンパツンです。表記のサイズはいい加減なので、１〜２cm小さいと考えてください。シャープのプラズマクラスター付きの現行品にテラスで使用のため、このカバーを買いましたが強風で上方にズレるため、引っ張って下に降ろす際に、すでに裏の結び紐の片側が切れてしまいました。簡単に切れるので気を付けたいですね？！ なので価格のことを考えると、６ヶ月（新年を迎えられれば）御の字ですが、果たして３ヶ月持つかどうか？様子見です。前のものがLEC（レック）という老舗のブランド（上場企業なのにアフターの対応も最悪なので２度とこのメーカーの製品は使わないと決めています）のもので１２００円位した割りには、１５ヶ月位しか持たなかったので、３〜４ヶ月使えればと。 →結果として１ヶ月でボロボロ。既に廃棄しました。"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87814cb5f60e0c844698", "gold_label": "ot"}
|
||||
{"id": "train-ja-003", "messages": [{"role": "user", "content": "一月過ぎたけどまだ来ない。最悪‼️流石中国星1つ付けたけど本当は星-5"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8781a95a610865f1e377", "gold_label": "ot"}
|
||||
{"id": "train-ja-004", "messages": [{"role": "user", "content": "埃に気を付け、湿度のある浴室で貼り付けを行いました。埃は問題が無かったのですが、残念ながら他の方も書かれていますが端の方に気泡が残ります、ガラス面と黒い額縁の境目に残ります。 爪で押し出しましたが気泡を追い出せません。 少し剥がし、気泡を追い出しましたがそういう問題で気泡が残っているわけでは無いようで、若干浮いていて気泡が入っている様です。 穴の部分はぴったりの位置にあるのでRの部分が相当シビアに作られているのかも知れません。 うまく貼れた場合でも、使用するとガラス面に結構指紋が付きます、何も貼らなかった場合に比べて滑らかさと防汚の面が劣りますが、これは全てのこういった商品に当てはまるのかも知れません。 ガラスの傷つきにくさですが、何も貼らなかった場合に比べて貼った方が強度は上のようです。 1ヵ月、XZ1をそのまま使用していると光に当てないと分からない位ですが、小傷が沢山出来ていました、こちらの商品を貼って1週間色々ハードに使いましたが傷は出来ておりません、強度は申し分ありません。"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87824d654f2ee1729b41", "gold_label": "ot"}
|
||||
{"id": "train-ja-005", "messages": [{"role": "user", "content": "マツエク休憩中で数年ぶりにマスカラを買いました。ダマにならず、しっかり付くのに本当にお湯でスルッと取れるので良いです！"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878215c0ba78d0881f34", "gold_label": "ot"}
|
||||
{"id": "train-ja-006", "messages": [{"role": "user", "content": "固定ができない･･･なんか斜めになる ズームするために回したら望遠本体まで動くから固定できず写真が撮れない。あと三脚なら固定できるだろうと思ってやったらできなかった…。魚眼レンズとかは全然使えるけど1番使いたかった望遠レンズが使えなくて残念。"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8782a5c09dd52de39324", "gold_label": "ot"}
|
||||
{"id": "train-ja-007", "messages": [{"role": "user", "content": "ずっと前から欲しかったですが、高価なイメージがあって。今回、充電の持ちと手頃なお値段で、こちらを選びました！ペアリングもスムーズにできたし、軽いし、耳へのフィット感も音もすごくいいです。ちなみに電子レンジ使う時だけ（特にレンジの使い始め）たまに途切れます。しかし、全然許容範囲です。"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8783bc7cc35e30f1df13", "gold_label": "ot"}
|
||||
{"id": "train-nl-000", "messages": [{"role": "user", "content": "Een hond die op een blikje bijt."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878389b4c0360c6466bf", "gold_label": "ot"}
|
||||
{"id": "train-nl-001", "messages": [{"role": "user", "content": "Rusland zegt dat ballistische raketten afgevuurd zijn in het Middellandse Zeegebied..."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878443d59830337427bb", "gold_label": "ot"}
|
||||
{"id": "train-nl-002", "messages": [{"role": "user", "content": "Iran atoombusbesprekingen beginnen in de hoop op vooruitgang..."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8783571e6ffa0b9caba9", "gold_label": "ot"}
|
||||
{"id": "train-nl-003", "messages": [{"role": "user", "content": "Twee vrouwen die op de bruine bank zitten."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8784e3dfcfab6fd1019e", "gold_label": "ot"}
|
||||
{"id": "train-nl-004", "messages": [{"role": "user", "content": "Er loopt een hond in de sneeuw."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87858b578eddec83028d", "gold_label": "ot"}
|
||||
{"id": "train-nl-005", "messages": [{"role": "user", "content": "Freddie Starr gearresteerd in Savile Abuse Sonde..."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8785ec03bada3b15cb48", "gold_label": "ot"}
|
||||
{"id": "train-nl-006", "messages": [{"role": "user", "content": "Een hond in een auto."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8785124dbf4a23c98923", "gold_label": "ot"}
|
||||
{"id": "train-nl-007", "messages": [{"role": "user", "content": "Een aanrecht en aanrechtblad met schalen op de planken."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878578b4e9389290f817", "gold_label": "ot"}
|
||||
{"id": "train-pl-000", "messages": [{"role": "user", "content": "Dwie dziewczyny z kucykami jeżdżą na przejażdżce w parku rozrywki."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8786a5f8e9319959d631", "gold_label": "ot"}
|
||||
{"id": "train-pl-001", "messages": [{"role": "user", "content": "alstom konkuruje o kontrakt z krajami japońskimi i niemieckimi."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8786d1bc7b6a3fb5f732", "gold_label": "ot"}
|
||||
{"id": "train-pl-002", "messages": [{"role": "user", "content": "Pankda je bambus."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8786f4fe3c88d68d6911", "gold_label": "ot"}
|
||||
{"id": "train-pl-003", "messages": [{"role": "user", "content": "Libijczycy zaczynają oddawać broń"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87865657fe29c59a3a6d", "gold_label": "ot"}
|
||||
{"id": "train-pl-004", "messages": [{"role": "user", "content": "4 zagranicznych żołnierzy zabitych na wschodzie Afganistanu"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8787c6a3424af9907964", "gold_label": "ot"}
|
||||
{"id": "train-pl-005", "messages": [{"role": "user", "content": "Egipt głosuje nad nową konstytucją"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8787bc7cc35e30f1df18", "gold_label": "ot"}
|
||||
{"id": "train-pl-006", "messages": [{"role": "user", "content": "Dwie osoby grają w golfa na polu golfowym."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8787d1bc7b6a3fb5f735", "gold_label": "ot"}
|
||||
{"id": "train-pl-007", "messages": [{"role": "user", "content": "Nigdy się z tego nie wydostaliśmy!"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87886aa29639cda94016", "gold_label": "ot"}
|
||||
{"id": "train-pt-000", "messages": [{"role": "user", "content": "Combatentes rebeldes 'capturam' soldados sírios"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8788dd32e5734ad628f4", "gold_label": "ot"}
|
||||
{"id": "train-pt-001", "messages": [{"role": "user", "content": "Nenhum acordo sobre penhasco fiscal enquanto Obama vai de férias"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878812e92b55023aa891", "gold_label": "ot"}
|
||||
{"id": "train-pt-002", "messages": [{"role": "user", "content": "Não foi possível contactar imediatamente representantes seqüenciais para comentários sobre o anúncio da SCO."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878931fa7ecf69e5f7fd", "gold_label": "ot"}
|
||||
{"id": "train-pt-003", "messages": [{"role": "user", "content": "Quatro aviões azuis e amarelos sobrevoando quatro barcos."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8789419b7910c2fd95f1", "gold_label": "ot"}
|
||||
{"id": "train-pt-004", "messages": [{"role": "user", "content": "O Comité Bancário do Senado está agendado para realizar uma audiência na terça-feira, onde Donaldson está agendado para testemunhar sobre hedge e fundos mútuos."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878b776710264cf67cde", "gold_label": "ot"}
|
||||
{"id": "train-pt-005", "messages": [{"role": "user", "content": "Thomas Cook acusado de colocar os custos à frente dos clientes"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878ab983a70a59cd3bf4", "gold_label": "ot"}
|
||||
{"id": "train-pt-006", "messages": [{"role": "user", "content": "Um cão corre à volta de um quintal."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878bc33554b59be5be88", "gold_label": "ot"}
|
||||
{"id": "train-pt-007", "messages": [{"role": "user", "content": "Um homem está a saltar uma parede."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878b55543c615b08135a", "gold_label": "ot"}
|
||||
{"id": "train-ru-000", "messages": [{"role": "user", "content": "Nsiad использует выводы таким образом в рамках своей текущей работы по двусторонним инициативам ."}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878a80beffa1ad578f7a", "gold_label": "ru"}
|
||||
{"id": "train-ru-001", "messages": [{"role": "user", "content": "Но это также о том времени , когда ты начинаешь получать температуру в кабине ."}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878c3393bfb96f2e6886", "gold_label": "ru"}
|
||||
{"id": "train-ru-002", "messages": [{"role": "user", "content": "Поощряется проведение экспериментов , равно как и представление такой дополнительной информации , что будет способствовать укреплению финансового доклада ."}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878c12e92b55023aa899", "gold_label": "ru"}
|
||||
{"id": "train-ru-003", "messages": [{"role": "user", "content": "Да , это совсем другая культура . Это странно внизу , потому что"}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878da770d98a4bce25ee", "gold_label": "ru"}
|
||||
{"id": "train-ru-004", "messages": [{"role": "user", "content": "Присяжные посмотрели вверх , заинтересовались ."}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878c697b682b78094377", "gold_label": "ru"}
|
||||
{"id": "train-ru-005", "messages": [{"role": "user", "content": "Но может ли он навсегда ?"}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878c5657fe29c59a3a7a", "gold_label": "ru"}
|
||||
{"id": "train-ru-006", "messages": [{"role": "user", "content": "И я могу назвать несколько больше , я просто не могу придумать их имена прямо с рук ."}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878ec85bf7c1921a6e62", "gold_label": "ru"}
|
||||
{"id": "train-ru-007", "messages": [{"role": "user", "content": "Межучрежденческой совет по делам бездомных находится на ранних этапах своего 10-летнего плана по ликвидации бездомности , как это было сказано ."}, {"role": "assistant", "content": "ru"}], "teacher_response_id": "chatcmpl-6a6a878ea5c09dd52de39331", "gold_label": "ru"}
|
||||
{"id": "train-sw-000", "messages": [{"role": "user", "content": "Isipokuwa kubwa ni uandishi wa habari wa ufaransa , ambapo heshima wakati kwa mitindo imekuwa ni mtazamo wa kawaida tangu karne ya 17"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878ea770d98a4bce25ef", "gold_label": "ot"}
|
||||
{"id": "train-sw-001", "messages": [{"role": "user", "content": "Mipangilio ya uga ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878eaff9f7f7f84cf451", "gold_label": "ot"}
|
||||
{"id": "train-sw-002", "messages": [{"role": "user", "content": "Ofisi ya usimamizi wa habari ( oim ) , ofisi ya utendaji wa programu ( opp ) , na timu ya kupanga ya serikali sasa ni wafanyakazi ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a878fe3dfcfab6fd101a9", "gold_label": "ot"}
|
||||
{"id": "train-sw-003", "messages": [{"role": "user", "content": "Nguvu hii ya asili ana na mwanaume tangu mara ya mapema , kama kuonyesha na pango-hekalu la pan , kwa ambaye raia wa syria na greeks kujitolea mkondo huo ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a879032ee9d9a487694f0", "gold_label": "ot"}
|
||||
{"id": "train-sw-004", "messages": [{"role": "user", "content": "Mawasiliano ya kibinafsi ( 2 ) pamoja na ande salimbot , undugu wa kimataifa wa boilermakers , wajenzi wa meli , blacksmiths , wazushi na wasaidizi , februari 22 , 2002 ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a879001c438c1be93d279", "gold_label": "ot"}
|
||||
{"id": "train-sw-005", "messages": [{"role": "user", "content": "Watu nani haki zao na kila kitu ng ' ambo na kila kitu na i think uh sijui kwa hiyo"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8790e5cd1f53d10ec53e", "gold_label": "ot"}
|
||||
{"id": "train-sw-006", "messages": [{"role": "user", "content": "Kuunda mazingira ya ushindani wange , kwa kiwango cha chini , kuondoa statutes4 ya kibinafsi na kanuni ya barua pepe ."}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8790f533e66c8281cad0", "gold_label": "ot"}
|
||||
{"id": "train-sw-007", "messages": [{"role": "user", "content": "Akiba ya kitaifa ya kitaifa"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87914c7866f47529520a", "gold_label": "ot"}
|
||||
{"id": "train-th-000", "messages": [{"role": "user", "content": "โอ้ มัน ฟัง ดู ยอดเยี่ยม"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87900b8d682056dbefad", "gold_label": "ot"}
|
||||
{"id": "train-th-001", "messages": [{"role": "user", "content": "ขอบใจ หลาย ๆ เด้อ รี เขา จากไป ด้วย การ เดิน แบบ เดียว กับ ที่ เขา มี จาก ซอย ที่ มั่นคง"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8791419b7910c2fd95f8", "gold_label": "ot"}
|
||||
{"id": "train-th-002", "messages": [{"role": "user", "content": "แต่ ฉัน เดา ว่า"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8791be759e5d2a00e630", "gold_label": "ot"}
|
||||
{"id": "train-th-003", "messages": [{"role": "user", "content": "ใช่ พวกเขา ควร ทำความสะอาด ให้ ชัดเจน ว่า มัน จะ ไม่ พา พวกเขา มาก ไป ใส่ สแตมป์ บน กระป๋อง น้ำผลไม้ ง่ายๆ เหมือน โซดา"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8791e3dfcfab6fd101ab", "gold_label": "ot"}
|
||||
{"id": "train-th-004", "messages": [{"role": "user", "content": "การ ตั้งครรภ์ ธรรมชาติ ได้รับ การ ยกระดับ โดย ระบบ ของ ยุโรป ที่ ล้ำสมัย ที่สุด ของ ชลบุรี ยังคง ดำเนินการ อยู่ ใน คลอง กลาง ที่ คุณ จะ เห็น บน ทาง ของ คุณ ใต้ ไป ยัง วี"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87924cb5f60e0c8446be", "gold_label": "ot"}
|
||||
{"id": "train-th-005", "messages": [{"role": "user", "content": "Atrisk งานก่อสร้าง ที่ แท้จริง มี การแสดง โดย ผู้รับเหมา การค้า ภายใต้ สัญญา กับ ซ. ผู้ แล้ว กลายเป็น ผู้รับผิดชอบ เจ้าของ สำหรับ การ ก่อสร้าง หมายถึง และ วิธีการ และ การจัดส่ง ของ สิ่งอำนวยความสะดวก ที่ สมบูรณ์ ภายใน ขอบเขต ของ เจ้าของงาน สำหรับ ต้นทุน เวลา และ คุณภาพ"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a87922d573972cd4cdff1", "gold_label": "ot"}
|
||||
{"id": "train-th-006", "messages": [{"role": "user", "content": "มัน ไม่ รบกวน ฉัน เลย ฉัน ไม่ รู้สึก ว่า มัน เป็นการ ละเมิด ความเป็นส่วนตัว หรือ อะไร เลย"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a879312e760f13afdab25", "gold_label": "ot"}
|
||||
{"id": "train-th-007", "messages": [{"role": "user", "content": "นี่ คือ ผล งานชิ้นเอก ของ อารยธรรม ฝรั่งเศส ที่ มี รอย เปลี่ยน ใน ศตวรรษ ที่ 12 จาก สไตล์ โร มาน ส ์ แบบ ไม่มีสติ ของ โบสถ์ เริ่ม สู่ ความ แข็งแกร่ง มากขึ้น"}, {"role": "assistant", "content": "ot"}], "teacher_response_id": "chatcmpl-6a6a8793adf8be004bbe4b3d", "gold_label": "ot"}
|
||||
{"id": "train-tr-000", "messages": [{"role": "user", "content": "Ayda 20 zloti veya her ay 20 zloti taksitle uygun bir alışveriş seçeneği sunuyoruz ."}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a8793a770d98a4bce25f4", "gold_label": "tr"}
|
||||
{"id": "train-tr-001", "messages": [{"role": "user", "content": "Amerika ' nın zamanında geleceğini içgüdüsel olarak hissetti ."}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a8794fabdd4e38cb710e0", "gold_label": "tr"}
|
||||
{"id": "train-tr-002", "messages": [{"role": "user", "content": "Santorini ve mykonos özellikle avrupa ve ABD ' den bir sürü tasarımcı giyim ve ayakkabı var ."}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a87944d654f2ee1729b50", "gold_label": "tr"}
|
||||
{"id": "train-tr-003", "messages": [{"role": "user", "content": "Ca ' daan , kara sürücüler karşılığında bir şey söyledi mi diye olabilir ."}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a87944c7866f47529520f", "gold_label": "tr"}
|
||||
{"id": "train-tr-004", "messages": [{"role": "user", "content": "Hareketleri dikkatlice ölçülür ."}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a8794a5c09dd52de39336", "gold_label": "tr"}
|
||||
{"id": "train-tr-005", "messages": [{"role": "user", "content": "Işte bunu söylediğin için mutluyum ve gerek yoktu ama o da aynı şekilde hissediyorum ben oturup bazı programları izliyorum ve uh ve bile dahil olan insanlar için utanıyorum bunu kendine neden yapıyorsun biliyorsun"}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a8795079009943e6a64fe", "gold_label": "tr"}
|
||||
{"id": "train-tr-006", "messages": [{"role": "user", "content": "Kazanan giriş , rock kendini"}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a87943f3c9b5f06007f13", "gold_label": "tr"}
|
||||
{"id": "train-tr-007", "messages": [{"role": "user", "content": "Örneğin , bir bilgisayar sisteminin güvenli olduğu ve tanımlanmış bir ortamda çalışmasına izin veren yazılı bir yetkilendirme ."}, {"role": "assistant", "content": "tr"}], "teacher_response_id": "chatcmpl-6a6a879412e92b55023aa8b2", "gold_label": "tr"}
|
||||
{"id": "train-ur-000", "messages": [{"role": "user", "content": "ماس ماس ( ماس ) نیٹ ورک کی بنیاد پر یہ اثر انداز کیا گیا ہے ."}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a87958d20c5ab26768723", "gold_label": "ur"}
|
||||
{"id": "train-ur-001", "messages": [{"role": "user", "content": "اس میں تاخیر کی جاتی ہے . \""}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a8797708f7e2c9beb0097", "gold_label": "ur"}
|
||||
{"id": "train-ur-002", "messages": [{"role": "user", "content": "اور ہم نے اس وقت تک ایک مدت معین کیا ہے جس کی وجہ سے اس سے ملاقات کی جاتی ہے تو ہم نے رات کے اوقات میں رات یا اس سے بھی کچھ کم نہیں کیا ۔ کوفتے کی تراکیب"}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a8797de34a86cafee51e1", "gold_label": "ur"}
|
||||
{"id": "train-ur-003", "messages": [{"role": "user", "content": "پھر ہم نے وزن کیا اور ہم نے اس میں سے ہر ایک کو عاجز کر دیا اور ہم ان سے بیزار ہیں"}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a87975657fe29c59a3a90", "gold_label": "ur"}
|
||||
{"id": "train-ur-004", "messages": [{"role": "user", "content": "کچھ بھی نہیں ۔"}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a8797448d84fdba942cd4", "gold_label": "ur"}
|
||||
{"id": "train-ur-005", "messages": [{"role": "user", "content": "جب میں نے آخری بار ایک سال میں پہلی بار ایسوسی ایشن سے اظہار کیا ، جس کی وجہ سے آپ نے ایک دوسرے کے پیمانے پر کام کرنے کی کوشش کی ، اس کے بعد ، دیگر ٹرانسپورٹ فرمیں کے ساتھ کام کرنے کے لئے ."}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a8798c683626717c42351", "gold_label": "ur"}
|
||||
{"id": "train-ur-006", "messages": [{"role": "user", "content": "یا آپ کا خیال ہے کہ آپ کو اس طرح کے کم از کم ایک بار نظر آنا چاہتے ہیں ؟"}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a87970b8d682056dbefb6", "gold_label": "ur"}
|
||||
{"id": "train-ur-007", "messages": [{"role": "user", "content": "واقعی رئیل اسپورٹس"}, {"role": "assistant", "content": "ur"}], "teacher_response_id": "chatcmpl-6a6a8798953d559c44153a94", "gold_label": "ur"}
|
||||
{"id": "train-vi-000", "messages": [{"role": "user", "content": "Một ứng cử viên alpha sẽ không cần sói chút nào cả ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a87984cff86af330ad15f", "gold_label": "vi"}
|
||||
{"id": "train-vi-001", "messages": [{"role": "user", "content": "Cô ấy là một người phụ nữ đáng chú ý , tăng sức mạnh làm nhiếp chính cho những thanh niên tutmosis ii con trai của cô ấy trước khi lấy nó cho chính mình bằng cách tự xưng là quyền thống trị ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a8799ad3e25d63f2c7976", "gold_label": "vi"}
|
||||
{"id": "train-vi-002", "messages": [{"role": "user", "content": "Những sự kiện đặc biệt như thú cưng cho thấy , thử thách chó , và những ngày thái lịch sử được tổ chức trong suốt mùa hề ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a879901c438c1be93d289", "gold_label": "vi"}
|
||||
{"id": "train-vi-003", "messages": [{"role": "user", "content": "Bây giờ hãy nhìn vào số tiền mà chính phủ có thể tiết kiệm nếu họ không có tất cả những ngày nghỉ trong những ngày nghỉ đó ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a8799a5f8e9319959d65a", "gold_label": "vi"}
|
||||
{"id": "train-vi-004", "messages": [{"role": "user", "content": "The Minato Mirai 21 dự án , được khởi chạy vào giữa những năm 1980 , đã được dự định sẽ biến một đường rộng lớn của bờ sông phía bắc và phía đông của sakuragi-cho vào một thành phố mô hình của tương lai , tích hợp kinh doanh , triển lãm , và giải trí cơ sở ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a879b166409dfd708b61a", "gold_label": "vi"}
|
||||
{"id": "train-vi-005", "messages": [{"role": "user", "content": "Tiêu chuẩn quốc tế một số công ty lớn / công ty e đã bảo vệ 9000 chứng nhận là một tổ chức chất lượng ( ISO ) 9000 hoạt động kiểm soát ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a879bfd4fe8c2d385a2d2", "gold_label": "vi"}
|
||||
{"id": "train-vi-006", "messages": [{"role": "user", "content": "Tất cả các bạn sẽ đến ăn tối với tôi ở savoy ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a879b166409dfd708b61c", "gold_label": "vi"}
|
||||
{"id": "train-vi-007", "messages": [{"role": "user", "content": "Ca ' daan ' không thể thấy được nếu những kỵ sĩ đen nói bất cứ điều gì để trở lại ."}, {"role": "assistant", "content": "vi"}], "teacher_response_id": "chatcmpl-6a6a879cf1ac6b519f479432", "gold_label": "vi"}
|
||||
{"id": "train-zh-000", "messages": [{"role": "user", "content": "东西不错，携带方便。但中国人鼻子比较低的，用的时候容易掉下来。"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879dc3316ea8a34aa6ef", "gold_label": "zh"}
|
||||
{"id": "train-zh-001", "messages": [{"role": "user", "content": "商品预测还有两周送货，我就提前申请退货取消订单，但是还是送过来了，联系客服怎么处理？客服说海外购订单无法取消，呵呵，买了就不许退，就这么硬气，然后退货必须自己出运费，什么玩意，垃圾商家亚马逊"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879d407571c0f21de54e", "gold_label": "zh"}
|
||||
{"id": "train-zh-002", "messages": [{"role": "user", "content": "本书共计368页，前57页包括解释全文，答记者问，理解适用，逐条释义，57页到165页案例，剩下部分都是法条，这么高定价，太黑了，请各位参考购买"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879db9c19ffa82a80169", "gold_label": "zh"}
|
||||
{"id": "train-zh-003", "messages": [{"role": "user", "content": "今天收到衣服， 很感人的卡片，真是体现的服务。超赞！"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879d4d654f2ee1729b56", "gold_label": "zh"}
|
||||
{"id": "train-zh-004", "messages": [{"role": "user", "content": "2016年4月11日下午4：31收到的天美时手表打开发现手表直接用表带绑在表盒内的塑料支架上，没有任何贴膜等保护措施，表带又轻又软质感很差，随后当天申请退货，4月12日上午顺丰快递寄回客服发的亚马逊钟表/珠宝频道，4月14日接到天津亚马逊打来电话，说经亚马逊检验手表没有贴膜且有磨损不符合退货要求。这就奇怪了 ，为什么寄出的时候根本不存在的贴膜，在退货检验的时候就变成理由了呢；为什么寄出的时候没有任何保护措施且直接绑在一个硬塑料支架上的这种包装方式反倒成了退货时手表有痕迹拒绝退货的理由了；而且网站的包装清单上没有注明任何关于这款手表应该有贴膜信息，如果说新品真的包含贴膜，那只能说明亚马逊将二手货和新品混合出售，这样作为消费者根本无法保护自己的权益；现在就造成亚马逊随便找一个理由就可以把这种二手产品直接推给消费者，这难道不属于变相的欺诈么。在此提醒其他想要购买亚马逊产品的消费者，至少中国亚马逊是没有道德底线的，高价商品千万不能在这购买。4月14日将这个情况反映到亚马逊之后，客服说会备注信息向上反映，4月16日再联系亚马逊时，被告知没有任何人员跟踪此事，亚马逊又开始采取拖延战术了。"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879da95a610865f1e393", "gold_label": "zh"}
|
||||
{"id": "train-zh-005", "messages": [{"role": "user", "content": "清仓的东西包装就这样 盒子发来就是开的还不是我打开的。不过东西不错啊 应该是全新没问题"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879efd361b86e4229935", "gold_label": "zh"}
|
||||
{"id": "train-zh-006", "messages": [{"role": "user", "content": "是我在亚马逊够买最不满意的物品，为什么换了这型号芯，出来的水烧开后有水渍，反而自来水却没有。为什么？厂家能解释吗？"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879e908217277579889a", "gold_label": "zh"}
|
||||
{"id": "train-zh-007", "messages": [{"role": "user", "content": "这车的价格太贵了，以为这个价格可以买好点的玩具，一打开包装我也是醉了，有种被骗的心"}, {"role": "assistant", "content": "zh"}], "teacher_response_id": "chatcmpl-6a6a879f8daee05e748dc32b", "gold_label": "zh"}
|
||||
+80
@@ -0,0 +1,80 @@
|
||||
{"id": "test-ar-000", "gold_label": "ar", "source_label": "ar", "prediction": "ar", "response": "ar", "correct": true, "input_tokens": 114, "output_tokens": 2, "latency_seconds": 0.021567012183368206}
|
||||
{"id": "test-ar-001", "gold_label": "ar", "source_label": "ar", "prediction": "ar", "response": "ar", "correct": true, "input_tokens": 156, "output_tokens": 2, "latency_seconds": 0.021308219991624355}
|
||||
{"id": "test-ar-002", "gold_label": "ar", "source_label": "ar", "prediction": "ar", "response": "ar", "correct": true, "input_tokens": 156, "output_tokens": 2, "latency_seconds": 0.021030287258327007}
|
||||
{"id": "test-ar-003", "gold_label": "ar", "source_label": "ar", "prediction": "ar", "response": "ar", "correct": true, "input_tokens": 91, "output_tokens": 2, "latency_seconds": 0.020903524942696095}
|
||||
{"id": "test-bg-000", "gold_label": "ot", "source_label": "bg", "prediction": "ru", "response": "ru", "correct": false, "input_tokens": 147, "output_tokens": 2, "latency_seconds": 0.020970353856682777}
|
||||
{"id": "test-bg-001", "gold_label": "ot", "source_label": "bg", "prediction": "ru", "response": "ru", "correct": false, "input_tokens": 69, "output_tokens": 2, "latency_seconds": 0.021014241501688957}
|
||||
{"id": "test-bg-002", "gold_label": "ot", "source_label": "bg", "prediction": "ru", "response": "ru", "correct": false, "input_tokens": 118, "output_tokens": 2, "latency_seconds": 0.021321195177733898}
|
||||
{"id": "test-bg-003", "gold_label": "ot", "source_label": "bg", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 85, "output_tokens": 2, "latency_seconds": 0.020931887440383434}
|
||||
{"id": "test-de-000", "gold_label": "de", "source_label": "de", "prediction": "de", "response": "de", "correct": true, "input_tokens": 174, "output_tokens": 2, "latency_seconds": 0.021052807569503784}
|
||||
{"id": "test-de-001", "gold_label": "de", "source_label": "de", "prediction": "de", "response": "de", "correct": true, "input_tokens": 43, "output_tokens": 2, "latency_seconds": 0.021180440671741962}
|
||||
{"id": "test-de-002", "gold_label": "de", "source_label": "de", "prediction": "de", "response": "de", "correct": true, "input_tokens": 60, "output_tokens": 2, "latency_seconds": 0.02106414269655943}
|
||||
{"id": "test-de-003", "gold_label": "de", "source_label": "de", "prediction": "de", "response": "de", "correct": true, "input_tokens": 43, "output_tokens": 2, "latency_seconds": 0.021330668590962887}
|
||||
{"id": "test-el-000", "gold_label": "el", "source_label": "el", "prediction": "el", "response": "el", "correct": true, "input_tokens": 165, "output_tokens": 2, "latency_seconds": 0.021219544112682343}
|
||||
{"id": "test-el-001", "gold_label": "el", "source_label": "el", "prediction": "el", "response": "el", "correct": true, "input_tokens": 111, "output_tokens": 2, "latency_seconds": 0.020995447412133217}
|
||||
{"id": "test-el-002", "gold_label": "el", "source_label": "el", "prediction": "el", "response": "el", "correct": true, "input_tokens": 92, "output_tokens": 2, "latency_seconds": 0.020973561331629753}
|
||||
{"id": "test-el-003", "gold_label": "el", "source_label": "el", "prediction": "el", "response": "el", "correct": true, "input_tokens": 92, "output_tokens": 2, "latency_seconds": 0.02086208201944828}
|
||||
{"id": "test-en-000", "gold_label": "en", "source_label": "en", "prediction": "en", "response": "en", "correct": true, "input_tokens": 50, "output_tokens": 2, "latency_seconds": 0.021106958389282227}
|
||||
{"id": "test-en-001", "gold_label": "en", "source_label": "en", "prediction": "en", "response": "en", "correct": true, "input_tokens": 69, "output_tokens": 2, "latency_seconds": 0.021015582606196404}
|
||||
{"id": "test-en-002", "gold_label": "en", "source_label": "en", "prediction": "en", "response": "en", "correct": true, "input_tokens": 81, "output_tokens": 2, "latency_seconds": 0.020907645113766193}
|
||||
{"id": "test-en-003", "gold_label": "en", "source_label": "en", "prediction": "en", "response": "en", "correct": true, "input_tokens": 57, "output_tokens": 2, "latency_seconds": 0.020868214778602123}
|
||||
{"id": "test-es-000", "gold_label": "es", "source_label": "es", "prediction": "es", "response": "es", "correct": true, "input_tokens": 149, "output_tokens": 2, "latency_seconds": 0.021015141159296036}
|
||||
{"id": "test-es-001", "gold_label": "es", "source_label": "es", "prediction": "es", "response": "es", "correct": true, "input_tokens": 63, "output_tokens": 2, "latency_seconds": 0.020898624323308468}
|
||||
{"id": "test-es-002", "gold_label": "es", "source_label": "es", "prediction": "es", "response": "es", "correct": true, "input_tokens": 134, "output_tokens": 2, "latency_seconds": 0.020956451073288918}
|
||||
{"id": "test-es-003", "gold_label": "es", "source_label": "es", "prediction": "es", "response": "es", "correct": true, "input_tokens": 72, "output_tokens": 2, "latency_seconds": 0.020912745036184788}
|
||||
{"id": "test-fr-000", "gold_label": "fr", "source_label": "fr", "prediction": "fr", "response": "fr", "correct": true, "input_tokens": 71, "output_tokens": 2, "latency_seconds": 0.020891825668513775}
|
||||
{"id": "test-fr-001", "gold_label": "fr", "source_label": "fr", "prediction": "fr", "response": "fr", "correct": true, "input_tokens": 57, "output_tokens": 2, "latency_seconds": 0.020838186144828796}
|
||||
{"id": "test-fr-002", "gold_label": "fr", "source_label": "fr", "prediction": "fr", "response": "fr", "correct": true, "input_tokens": 85, "output_tokens": 2, "latency_seconds": 0.02096572984009981}
|
||||
{"id": "test-fr-003", "gold_label": "fr", "source_label": "fr", "prediction": "fr", "response": "fr", "correct": true, "input_tokens": 65, "output_tokens": 2, "latency_seconds": 0.020963875576853752}
|
||||
{"id": "test-hi-000", "gold_label": "hi", "source_label": "hi", "prediction": "hi", "response": "hi", "correct": true, "input_tokens": 86, "output_tokens": 2, "latency_seconds": 0.02084300108253956}
|
||||
{"id": "test-hi-001", "gold_label": "hi", "source_label": "hi", "prediction": "hi", "response": "hi", "correct": true, "input_tokens": 86, "output_tokens": 2, "latency_seconds": 0.020935524255037308}
|
||||
{"id": "test-hi-002", "gold_label": "hi", "source_label": "hi", "prediction": "hi", "response": "hi", "correct": true, "input_tokens": 153, "output_tokens": 2, "latency_seconds": 0.020963420160114765}
|
||||
{"id": "test-hi-003", "gold_label": "hi", "source_label": "hi", "prediction": "hi", "response": "hi", "correct": true, "input_tokens": 124, "output_tokens": 2, "latency_seconds": 0.02094507310539484}
|
||||
{"id": "test-it-000", "gold_label": "ot", "source_label": "it", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 44, "output_tokens": 2, "latency_seconds": 0.020852336660027504}
|
||||
{"id": "test-it-001", "gold_label": "ot", "source_label": "it", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 49, "output_tokens": 2, "latency_seconds": 0.020954391919076443}
|
||||
{"id": "test-it-002", "gold_label": "ot", "source_label": "it", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 96, "output_tokens": 2, "latency_seconds": 0.02085470873862505}
|
||||
{"id": "test-it-003", "gold_label": "ot", "source_label": "it", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 44, "output_tokens": 2, "latency_seconds": 0.0210230415686965}
|
||||
{"id": "test-ja-000", "gold_label": "ot", "source_label": "ja", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 193, "output_tokens": 2, "latency_seconds": 0.021084836684167385}
|
||||
{"id": "test-ja-001", "gold_label": "ot", "source_label": "ja", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 100, "output_tokens": 2, "latency_seconds": 0.020785780623555183}
|
||||
{"id": "test-ja-002", "gold_label": "ot", "source_label": "ja", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 121, "output_tokens": 2, "latency_seconds": 0.021001679822802544}
|
||||
{"id": "test-ja-003", "gold_label": "ot", "source_label": "ja", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 134, "output_tokens": 2, "latency_seconds": 0.020998574793338776}
|
||||
{"id": "test-nl-000", "gold_label": "ot", "source_label": "nl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 76, "output_tokens": 2, "latency_seconds": 0.020837283693253994}
|
||||
{"id": "test-nl-001", "gold_label": "ot", "source_label": "nl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 46, "output_tokens": 2, "latency_seconds": 0.020906995981931686}
|
||||
{"id": "test-nl-002", "gold_label": "ot", "source_label": "nl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 44, "output_tokens": 2, "latency_seconds": 0.02119743824005127}
|
||||
{"id": "test-nl-003", "gold_label": "ot", "source_label": "nl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 81, "output_tokens": 2, "latency_seconds": 0.020859770476818085}
|
||||
{"id": "test-pl-000", "gold_label": "ot", "source_label": "pl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 64, "output_tokens": 2, "latency_seconds": 0.0209045996889472}
|
||||
{"id": "test-pl-001", "gold_label": "ot", "source_label": "pl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 60, "output_tokens": 2, "latency_seconds": 0.02100309729576111}
|
||||
{"id": "test-pl-002", "gold_label": "ot", "source_label": "pl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 71, "output_tokens": 2, "latency_seconds": 0.020983186550438404}
|
||||
{"id": "test-pl-003", "gold_label": "ot", "source_label": "pl", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 90, "output_tokens": 2, "latency_seconds": 0.020953955128788948}
|
||||
{"id": "test-pt-000", "gold_label": "ot", "source_label": "pt", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 62, "output_tokens": 2, "latency_seconds": 0.02100022416561842}
|
||||
{"id": "test-pt-001", "gold_label": "ot", "source_label": "pt", "prediction": "es", "response": "es", "correct": false, "input_tokens": 51, "output_tokens": 2, "latency_seconds": 0.021064297296106815}
|
||||
{"id": "test-pt-002", "gold_label": "ot", "source_label": "pt", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 46, "output_tokens": 2, "latency_seconds": 0.020892093889415264}
|
||||
{"id": "test-pt-003", "gold_label": "ot", "source_label": "pt", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 62, "output_tokens": 2, "latency_seconds": 0.020917534828186035}
|
||||
{"id": "test-ru-000", "gold_label": "ru", "source_label": "ru", "prediction": "ru", "response": "ru", "correct": true, "input_tokens": 94, "output_tokens": 2, "latency_seconds": 0.020882914774119854}
|
||||
{"id": "test-ru-001", "gold_label": "ru", "source_label": "ru", "prediction": "ru", "response": "ru", "correct": true, "input_tokens": 55, "output_tokens": 2, "latency_seconds": 0.02089247014373541}
|
||||
{"id": "test-ru-002", "gold_label": "ru", "source_label": "ru", "prediction": "ru", "response": "ru", "correct": true, "input_tokens": 55, "output_tokens": 2, "latency_seconds": 0.02086214069277048}
|
||||
{"id": "test-ru-003", "gold_label": "ru", "source_label": "ru", "prediction": "ru", "response": "ru", "correct": true, "input_tokens": 104, "output_tokens": 2, "latency_seconds": 0.020866707898676395}
|
||||
{"id": "test-sw-000", "gold_label": "ot", "source_label": "sw", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 101, "output_tokens": 2, "latency_seconds": 0.02101822104305029}
|
||||
{"id": "test-sw-001", "gold_label": "ot", "source_label": "sw", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 97, "output_tokens": 2, "latency_seconds": 0.02105732262134552}
|
||||
{"id": "test-sw-002", "gold_label": "ot", "source_label": "sw", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 80, "output_tokens": 2, "latency_seconds": 0.020854887552559376}
|
||||
{"id": "test-sw-003", "gold_label": "ot", "source_label": "sw", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 73, "output_tokens": 2, "latency_seconds": 0.020936796441674232}
|
||||
{"id": "test-th-000", "gold_label": "ot", "source_label": "th", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 259, "output_tokens": 2, "latency_seconds": 0.02173076570034027}
|
||||
{"id": "test-th-001", "gold_label": "ot", "source_label": "th", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 257, "output_tokens": 2, "latency_seconds": 0.021232523024082184}
|
||||
{"id": "test-th-002", "gold_label": "ot", "source_label": "th", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 108, "output_tokens": 2, "latency_seconds": 0.020946724340319633}
|
||||
{"id": "test-th-003", "gold_label": "ot", "source_label": "th", "prediction": "ot", "response": "ot", "correct": true, "input_tokens": 122, "output_tokens": 2, "latency_seconds": 0.020870834589004517}
|
||||
{"id": "test-tr-000", "gold_label": "tr", "source_label": "tr", "prediction": "tr", "response": "tr", "correct": true, "input_tokens": 92, "output_tokens": 2, "latency_seconds": 0.02091517671942711}
|
||||
{"id": "test-tr-001", "gold_label": "tr", "source_label": "tr", "prediction": "tr", "response": "tr", "correct": true, "input_tokens": 92, "output_tokens": 2, "latency_seconds": 0.020825224928557873}
|
||||
{"id": "test-tr-002", "gold_label": "tr", "source_label": "tr", "prediction": "tr", "response": "tr", "correct": true, "input_tokens": 71, "output_tokens": 2, "latency_seconds": 0.02104836329817772}
|
||||
{"id": "test-tr-003", "gold_label": "tr", "source_label": "tr", "prediction": "tr", "response": "tr", "correct": true, "input_tokens": 120, "output_tokens": 2, "latency_seconds": 0.02091341372579336}
|
||||
{"id": "test-ur-000", "gold_label": "ur", "source_label": "ur", "prediction": "ur", "response": "ur", "correct": true, "input_tokens": 129, "output_tokens": 2, "latency_seconds": 0.02105330489575863}
|
||||
{"id": "test-ur-001", "gold_label": "ur", "source_label": "ur", "prediction": "ur", "response": "ur", "correct": true, "input_tokens": 117, "output_tokens": 2, "latency_seconds": 0.021038753911852837}
|
||||
{"id": "test-ur-002", "gold_label": "ur", "source_label": "ur", "prediction": "ur", "response": "ur", "correct": true, "input_tokens": 166, "output_tokens": 2, "latency_seconds": 0.02175528835505247}
|
||||
{"id": "test-ur-003", "gold_label": "ur", "source_label": "ur", "prediction": "ur", "response": "ur", "correct": true, "input_tokens": 64, "output_tokens": 2, "latency_seconds": 0.022008763626217842}
|
||||
{"id": "test-vi-000", "gold_label": "vi", "source_label": "vi", "prediction": "vi", "response": "vi", "correct": true, "input_tokens": 123, "output_tokens": 2, "latency_seconds": 0.02193110343068838}
|
||||
{"id": "test-vi-001", "gold_label": "vi", "source_label": "vi", "prediction": "vi", "response": "vi", "correct": true, "input_tokens": 176, "output_tokens": 2, "latency_seconds": 0.021714639849960804}
|
||||
{"id": "test-vi-002", "gold_label": "vi", "source_label": "vi", "prediction": "vi", "response": "vi", "correct": true, "input_tokens": 66, "output_tokens": 2, "latency_seconds": 0.021408547647297382}
|
||||
{"id": "test-vi-003", "gold_label": "vi", "source_label": "vi", "prediction": "vi", "response": "vi", "correct": true, "input_tokens": 103, "output_tokens": 2, "latency_seconds": 0.0215337872505188}
|
||||
{"id": "test-zh-000", "gold_label": "zh", "source_label": "zh", "prediction": "zh", "response": "zh", "correct": true, "input_tokens": 85, "output_tokens": 2, "latency_seconds": 0.02103542722761631}
|
||||
{"id": "test-zh-001", "gold_label": "zh", "source_label": "zh", "prediction": "zh", "response": "zh", "correct": true, "input_tokens": 109, "output_tokens": 2, "latency_seconds": 0.02097904495894909}
|
||||
{"id": "test-zh-002", "gold_label": "zh", "source_label": "zh", "prediction": "zh", "response": "zh", "correct": true, "input_tokens": 99, "output_tokens": 2, "latency_seconds": 0.02109034825116396}
|
||||
{"id": "test-zh-003", "gold_label": "zh", "source_label": "zh", "prediction": "zh", "response": "zh", "correct": true, "input_tokens": 69, "output_tokens": 2, "latency_seconds": 0.021113927476108074}
|
||||
+14
@@ -0,0 +1,14 @@
|
||||
{
|
||||
"arm": "trained",
|
||||
"rows": 80,
|
||||
"correct": 76,
|
||||
"accuracy": 0.95,
|
||||
"parse_rate": 1.0,
|
||||
"input_tokens": 7738,
|
||||
"output_tokens": 160,
|
||||
"latency_seconds_total": 1.6843466674908996,
|
||||
"latency_seconds_mean": 0.021054333343636246,
|
||||
"latency_seconds_median": 0.02097630314528942,
|
||||
"provider_charge_usd": 0.0,
|
||||
"provider_charge_note": "Local inference made no provider API calls; electricity/hardware amortization is excluded."
|
||||
}
|
||||
+240
@@ -0,0 +1,240 @@
|
||||
{"id": "train-ar-000", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "اسفل على اليمين مطعم ماكسيم , الذي بدا ك a صالون ايس كريم وهو الان a نصب اكثر الموقر من مادلين .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876c0b8d682056dbef7b", "response_model": "kimi-k3", "response_created": 1785366380, "request_started_at": "2026-07-29T23:06:19.575854+00:00", "latency_seconds": 4.317037624772638, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is predominantly Arabic script with some Latin letter \"a\" inserted oddly (likely artifacts). Arabic vocabulary: اسفل على اليمين مطعم... الذي بدا... صالون ايس كريم وهو الان... نصب اكثر الموقر من مادلين. This is Arabic despite weird Latin a's. Classify ar. Need final exact line only.", "prediction": "ar", "usage": {"completion_tokens": 102, "prompt_tokens": 389, "total_tokens": 491, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 83, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01798, "usd": 0.002657303896103896}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "40c59fde495c7d440f2f2ffc4e4639bc7932bf34d8aec305b9e3f2ee8ba3c89a"}, "error": null}
|
||||
{"id": "train-ar-001", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "هل تحب الافلام المعلقة او هل تحب فقط اكشن او", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876c3d093ac09861553b", "response_model": "kimi-k3", "response_created": 1785366380, "request_started_at": "2026-07-29T23:06:19.643649+00:00", "latency_seconds": 3.5656825830228627, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is Arabic script: \"هل تحب الافلام المعلقة او هل تحب فقط اكشن او\" meaning \"Do you like dubbed/commented films or do you like only action or\". Clearly Arabic. No Urdu-specific letters. Final answer ar.", "prediction": "ar", "usage": {"completion_tokens": 79, "prompt_tokens": 365, "total_tokens": 444, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 60, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.0152, "usd": 0.0022464415584415587}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "2168d75af3554e862e5113bb9434d0a9ddbba6ebc08ed292ed0f367caf098223"}, "error": null}
|
||||
{"id": "train-ar-002", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "اعاد الموقع فورا موقع برادلي الرسمي , موقع لاخبار عن السباق الرئاسي لعام 2000 , معلومات عن البطاقة الطبية , والكتب عن برادلي .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876c7d03af917e2edc1a", "response_model": "kimi-k3", "response_created": 1785366381, "request_started_at": "2026-07-29T23:06:19.644119+00:00", "latency_seconds": 4.169204874895513, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is in Arabic script. Words like \"اعاد الموقع\", \"فورا\", \"موقع برادلي الرسمي\", \"السباق الرئاسي\", etc. are Arabic, not Urdu (no Urdu-specific letters like ٹ ڈ ڑ ں ہ ھ). So label ar.", "prediction": "ar", "usage": {"completion_tokens": 88, "prompt_tokens": 401, "total_tokens": 489, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 69, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01682, "usd": 0.0024858649350649356}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "fe91216602b6e7c261208bfeae07469155ec676d169e71b6aa628f26f176943a"}, "error": null}
|
||||
{"id": "train-ar-003", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "هل انت مشترك في اي هندسة رسم الاشياء التي", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876cf4fe3c88d68d68fd", "response_model": "kimi-k3", "response_created": 1785366380, "request_started_at": "2026-07-29T23:06:19.644456+00:00", "latency_seconds": 4.23745295824483, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is Arabic script. Need distinguish Arabic vs Urdu. The phrase: \"هل انت مشترك في اي هندسة رسم الاشياء التي\" is clearly Arabic: هل انت مشترك في أي هندسة... uses Arabic words, no Urdu-specific letters (ے، ٹ، ڈ، ڑ etc) or Urdu vocabulary. Label ar. Need final exactly one line.", "prediction": "ar", "usage": {"completion_tokens": 106, "prompt_tokens": 363, "total_tokens": 469, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 87, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01786, "usd": 0.0026395688311688314}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "35d34cd248e91a7a3d2877c357d2d7e6b969999838d4161f0eac3f1786e8d9fa"}, "error": null}
|
||||
{"id": "train-ar-004", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "00-187-الاثار المترتبة على الضمان الاجتماعي بالنسبة للمعاشات التقاعدية الخاصة ( جاو / hehs-00-187 , 14 ايلول / سبتمبر 2000 ) .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876c571e6ffa0b9cab99", "response_model": "kimi-k3", "response_created": 1785366380, "request_started_at": "2026-07-29T23:06:19.647153+00:00", "latency_seconds": 3.2070549158379436, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text contains Arabic script with words like الاثار المترتبة على الضمان الاجتماعي etc. Clearly Arabic despite numbers and some Latin code hehs. Label ar. Need final exactly one line.", "prediction": "ar", "usage": {"completion_tokens": 66, "prompt_tokens": 405, "total_tokens": 471, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 47, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.0147, "usd": 0.0021725454545454548}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "467527b859adace4bc9a1d98e7cc1af8809aae5d71af9ad05534b8c077d9744d"}, "error": null}
|
||||
{"id": "train-ar-005", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "ويشمل موقع anao وصلات بمنشوراتها المختلفة , بما في ذلك تقارير مراجعة الحسابات و يهدي الممارسات الافضل .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876cb9c19ffa82a80140", "response_model": "kimi-k3", "response_created": 1785366381, "request_started_at": "2026-07-29T23:06:19.648452+00:00", "latency_seconds": 5.287996749859303, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is primarily Arabic script: ويشمل موقع ... وصلات بمنشوراتها المختلفة بما في ذلك تقارير مراجعة الحسابات و يهدي الممارسات الافضل. Contains Latin 'anao' proper noun/acronym, ignore/proper noun. Arabic vocabulary clear. Not Urdu (no Urdu-specific letters like ے ٹ ڈ ڑ and words). Classify ar.", "prediction": "ar", "usage": {"completion_tokens": 114, "prompt_tokens": 391, "total_tokens": 505, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 95, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01922, "usd": 0.0028405662337662343}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "dae3f030ebc2757205fd072a6b9fc36d1e40f42f91297400a321044807b820ff"}, "error": null}
|
||||
{"id": "train-ar-006", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "شاليت يقول انه عندما تمشي في الشارع يمكنك ان تخبر العذارى بسبب وهجهم الطازج والمتوازن .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876cde3406f5dc3ef752", "response_model": "kimi-k3", "response_created": 1785366381, "request_started_at": "2026-07-29T23:06:19.648742+00:00", "latency_seconds": 5.195527957752347, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The user wants me to classify the language of the text. The text is in Arabic script. It reads: \"شاليت يقول انه عندما تمشي في الشارع يمكنك ان تخبر العذارى بسبب وهجهم الطازج والمتوازن .\" This is clearly Arabic. No Urdu-specific letters like ے، ٹ، ڈ، ڑ. The vocabulary is Arabic (يقول، انه، عندما، تمشي، الشارع، يمكنك). Label ar. Need final exactly one line.", "prediction": "ar", "usage": {"completion_tokens": 145, "prompt_tokens": 388, "total_tokens": 533, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 126, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.02226, "usd": 0.0032898545454545457}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "4d1f48ed59bf07bc591011aff42be47c53a7ce3144e89bd90ff10f8869892e89"}, "error": null}
|
||||
{"id": "train-ar-007", "split": "train", "source_label": "ar", "gold_label": "ar", "text": "- ويكاد يكون تصنيع حافزا حافزا مكرسا تماما لتطبيقات توليد الطاقة .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876cde3406f5dc3ef753", "response_model": "kimi-k3", "response_created": 1785366381, "request_started_at": "2026-07-29T23:06:19.649086+00:00", "latency_seconds": 3.0407175407744944, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "Text Arabic script: ويكاد يكون تصنيع حافزا ... Arabic vocabulary. No Urdu-specific letters. Label ar. Need final exact only.", "prediction": "ar", "usage": {"completion_tokens": 53, "prompt_tokens": 378, "total_tokens": 431, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 34, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01286, "usd": 0.0019006077922077923}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "233fdb01e5700148eafe4419b90f576277cb6b4960c5d56db22b1c22ef779740"}, "error": null}
|
||||
{"id": "train-bg-000", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "това е на описание на г-н браун !", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876cbc7cc35e30f1df0a", "response_model": "kimi-k3", "response_created": 1785366380, "request_started_at": "2026-07-29T23:06:19.651000+00:00", "latency_seconds": 3.8397959996946156, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text uses Cyrillic script. It says \"това е на описание на г-н браун !\" which is Bulgarian (\"this is a description of Mr. Brown\"). Bulgarian is not among the listed labels (ru is Russian, but this is Bulgarian vocabulary: това е, описание, г-н). So classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 97, "prompt_tokens": 358, "total_tokens": 455, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 78, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01686, "usd": 0.0024917766233766236}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f1f395eb643acaea78c37cd9698e8ed348bc79ee4636417ce9bd3ac7e3c631b1"}, "error": null}
|
||||
{"id": "train-bg-001", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "за финансовата 2002 година vba преразгледа плановете си за ефективност на висшите ръководители в регионалните служби за подобряване на индивидуалната отчетност за елементите на изпълнение чрез свързване на целите на организацията за ефективност и действителните резултати с значими и измерими елементи на ефективност .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876ca5c09dd52de3930e", "response_model": "kimi-k3", "response_created": 1785366381, "request_started_at": "2026-07-29T23:06:19.652264+00:00", "latency_seconds": 2.169222625438124, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Bulgarian → ot.", "prediction": "ot", "usage": {"completion_tokens": 26, "prompt_tokens": 470, "total_tokens": 496, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 7, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.012, "usd": 0.0017735064935064937}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7e000f129f49bc9cf907a62a784aa1be6f94396ac56db10824ab36277fc5eb1d"}, "error": null}
|
||||
{"id": "train-bg-002", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "журналистът е измамник , предназначен само да кара трафика до порно сайтовете на собственика .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876cccc0dbe1b98e03c1", "response_model": "kimi-k3", "response_created": 1785366380, "request_started_at": "2026-07-29T23:06:19.652922+00:00", "latency_seconds": 5.368339417036623, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Cyrillic script. Words: \"журналистът е измамник, предназначен само да кара трафика до порно сайтовете на собственика.\" This looks Bulgarian, not Russian. Bulgarian uses definite article suffix -ът (журналистът), \"е\", \"да\", \"на\". Russian would be \"журналист — мошенник\" etc. Bulgarian is not in listed labels, so classify as ot. Need final exactly.", "prediction": "ot", "usage": {"completion_tokens": 130, "prompt_tokens": 379, "total_tokens": 509, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 111, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.02058, "usd": 0.003041563636363637}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "ad3648cb9763c85b40074df3e34dc19731e4ff8bbf10fe658de245d39a0dd437"}, "error": null}
|
||||
{"id": "train-bg-003", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "толкова добре , колкото хората казват , че е .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876cb9c19ffa82a8013f", "response_model": "kimi-k3", "response_created": 1785366381, "request_started_at": "2026-07-29T23:06:19.653228+00:00", "latency_seconds": 3.7638345840387046, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Cyrillic: \"толкова добре, колкото хората казват, че е.\" This is Bulgarian (\"as good as people say it is\"). Bulgarian is not in the listed labels, so it should be classified as ot (all other languages). Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 86, "prompt_tokens": 365, "total_tokens": 451, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 67, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.0159, "usd": 0.002349896103896104}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "07140648c35de6091b539b3464ebb046d2840e8de285cac2af466336d9bb20d3"}, "error": null}
|
||||
{"id": "train-bg-004", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "о , имаш ли такова време в северна каролина ?", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876eb9c19ffa82a80142", "response_model": "kimi-k3", "response_created": 1785366383, "request_started_at": "2026-07-29T23:06:21.822380+00:00", "latency_seconds": 4.722192083951086, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text uses Cyrillic script. It reads \"о , имаш ли такова време в северна каролина ?\" This is Bulgarian, meaning \"oh, do you have such weather in North Carolina?\" Bulgarian is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh), so it should be classified as ot (other). Not Russian — Bulgarian has \"имаш ли\", \"такова\", vocabulary distinct from Russian. Final answer ot.", "prediction": "ot", "usage": {"completion_tokens": 132, "prompt_tokens": 363, "total_tokens": 495, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 113, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015852, "usd": 0.0023428020779220782}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7657a6aa267dec4509ddca3790fadf55de964f9fc32349469c667208dd3f3571"}, "error": null}
|
||||
{"id": "train-bg-005", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "както и да е , закуската е била принуден в тях сред жените преди известно време , така че няма за какво да се тревожим .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876fd1bc7b6a3fb5f715", "response_model": "kimi-k3", "response_created": 1785366384, "request_started_at": "2026-07-29T23:06:22.690708+00:00", "latency_seconds": 5.540165749844164, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Cyrillic. Words: \"както и да е, закуската е била принуден в тях сред жените преди известно време, така че няма за какво да се тревожим.\" This is Bulgarian, not Russian. Bulgarian uses Cyrillic but vocabulary/grammar differs: \"както и да е\", \"закуската\", \"няма за какво да се тревожим\" are Bulgarian. Listed labels include ru but not bg; other languages -> ot. Need final exactly.", "prediction": "ot", "usage": {"completion_tokens": 146, "prompt_tokens": 395, "total_tokens": 541, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 127, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.0225, "usd": 0.0033253246753246757}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "87daf1c42ae456d43103db18cf4fe7da53e19e3125af2494449f0297df2dbd07"}, "error": null}
|
||||
{"id": "train-bg-006", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "тийнейджърите биха искали да се присъединят към на диско орди в хараджуку , близо до парка yoyogi .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876f419b7910c2fd95d3", "response_model": "kimi-k3", "response_created": 1785366384, "request_started_at": "2026-07-29T23:06:22.855093+00:00", "latency_seconds": 2.906798125244677, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Bulgarian Cyrillic, not Russian. Label ot.", "prediction": "ot", "usage": {"completion_tokens": 32, "prompt_tokens": 388, "total_tokens": 420, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 13, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01096, "usd": 0.0016198025974025975}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "6f5e76893f3b796f9dc6d6fe6814e5bc20e8320d15fac96127fe9af232e6d961"}, "error": null}
|
||||
{"id": "train-bg-007", "split": "train", "source_label": "bg", "gold_label": "ot", "text": "циници стенеше , че хитчинс се е цел в клинтън , но вместо това е застрелял блументал .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a876f7d03af917e2edc1d", "response_model": "kimi-k3", "response_created": 1785366384, "request_started_at": "2026-07-29T23:06:23.210318+00:00", "latency_seconds": 6.40476637519896, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Cyrillic: \"циници стенеше , че хитчинс се е цел в клинтън , но вместо това е застрелял блументал .\" This looks Bulgarian (Macedonian possible). Words: циници (cynics), стенеше, че, се е цел? \"се е цел в\" maybe \"aimed at\"; но вместо това (but instead), застрелял (shot). It's not Russian (no ъ, uses Bulgarian-specific е/ъ and grammar), outside listed labels, Cyrillic not Russian. Label ot. Need final exact line.", "prediction": "ot", "usage": {"completion_tokens": 166, "prompt_tokens": 385, "total_tokens": 551, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 147, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.019692, "usd": 0.0029103241558441563}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0e808043609cc4ff4119b5acab2c061ccc25fbdc3ffa3c1484080f79c9c6a642"}, "error": null}
|
||||
{"id": "train-de-000", "split": "train", "source_label": "de", "gold_label": "de", "text": "Sieht gut aus und hat super an meinen skoda Schlüssel gepasst", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87708ff452ff67d52a45", "response_model": "kimi-k3", "response_created": 1785366385, "request_started_at": "2026-07-29T23:06:23.417895+00:00", "latency_seconds": 4.473418999928981, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The user wants a classification only. Need determine language. Text: \"Sieht gut aus und hat super an meinen skoda Schlüssel gepasst\". German words: Sieht gut aus = looks good, und, hat super an meinen ... Schlüssel gepasst. Clearly German. Need output exactly one line Final Answer: de. Ensure no extra.", "prediction": "de", "usage": {"completion_tokens": 98, "prompt_tokens": 364, "total_tokens": 462, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 79, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01708, "usd": 0.0025242909090909095}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "49b89356198aa74a8a47ee2f46c1347efcf8123368978d05c727d9b24f9cc106"}, "error": null}
|
||||
{"id": "train-de-001", "split": "train", "source_label": "de", "gold_label": "de", "text": "Kann mich obenstehenden Rezensionen nur anschließen, das Headset ist im Grunde genommen sein Geld wert, würde es nicht nach einem Jahr mit Wackelkontakt den Geist aufgeben.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87705a513bb219e7134f", "response_model": "kimi-k3", "response_created": 1785366385, "request_started_at": "2026-07-29T23:06:23.491600+00:00", "latency_seconds": 4.076920416671783, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text is clearly German: \"Kann mich obenstehenden Rezensionen nur anschließen...\" contains German vocabulary and the ß character. Label de. Need output exactly Final Answer: de.", "prediction": "de", "usage": {"completion_tokens": 66, "prompt_tokens": 399, "total_tokens": 465, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 47, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009972, "usd": 0.0014737838961038962}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f6c1481da7665408126cc30d029cc66123735fed44e91144d97cc1974216d51a"}, "error": null}
|
||||
{"id": "train-de-002", "split": "train", "source_label": "de", "gold_label": "de", "text": "Eine Flasche ist leider mehr oder weniger ausgelaufen. Nichtsdestotrotz tut die Kochsalzlösung was sie soll. Punktabzug gibt es nur wegen dem auslaufen.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87703d093ac09861553e", "response_model": "kimi-k3", "response_created": 1785366384, "request_started_at": "2026-07-29T23:06:23.813928+00:00", "latency_seconds": 2.6244437908753753, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text is clearly German: \"Eine Flasche ist leider mehr oder weniger ausgelaufen...\" contains German vocabulary and function words. Need final exactly format.", "prediction": "de", "usage": {"completion_tokens": 55, "prompt_tokens": 393, "total_tokens": 448, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 36, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008752, "usd": 0.0012934774025974026}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "bb91888e9f9826620d9fbfe8c223f5603ae66ba3bac311d9d263560c4cffb59b"}, "error": null}
|
||||
{"id": "train-de-003", "split": "train", "source_label": "de", "gold_label": "de", "text": "Habe die Tasche als Geschenk gekauft - Lieferung funktionierte spitze und die Tasche sieht wirklich toll aus (Verarbeitung, Stoff usw.! Daumen hoch!", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8770bcb8b1ca0dc259ad", "response_model": "kimi-k3", "response_created": 1785366385, "request_started_at": "2026-07-29T23:06:23.882548+00:00", "latency_seconds": 3.7766718328930438, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text is German: \"Habe die Tasche als Geschenk gekauft - Lieferung funktionierte spitze und die Tasche sieht wirklich toll aus (Verarbeitung, Stoff usw.! Daumen hoch!\" Clear German vocabulary and function words. Final label de.", "prediction": "de", "usage": {"completion_tokens": 89, "prompt_tokens": 396, "total_tokens": 485, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 70, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01682, "usd": 0.0024858649350649356}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "106e63a8c819f2934c98ba7bab51004ef3c35087ccedcb05f2adfe8effb3e6ec"}, "error": null}
|
||||
{"id": "train-de-004", "split": "train", "source_label": "de", "gold_label": "de", "text": "Gut finde ich die Optik und die Ersatzbürsten. Nicht ganz durchdacht finde ich den Behälter, wo die Klobürste rein kommt. Nimmt man diese raus und steckt sie wieder ein, reibt der Behälter über den Fliesenboden (den man üblicherweise in der Toilette hat) was was Geräuche macht und einen schlechten Qualitätseindruck hinterlässt. Sinnvoll wäre, wenn auf der Unterseite noch eine Gummifläche geklebt würde, so dass Metall nicht auf Fliesen reibt.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8770a5c09dd52de39312", "response_model": "kimi-k3", "response_created": 1785366384, "request_started_at": "2026-07-29T23:06:23.893816+00:00", "latency_seconds": 3.6658056252636015, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The user wants language classification. Text is clearly German: \"Gut finde ich die Optik... Ersatzbürsten... Behälter... Klobürste... Fliesenboden... Toilette... Geräusche... Qualitätseindruck...\" Need output exactly one line Final Answer: de. Ensure no extra. ", "prediction": "de", "usage": {"completion_tokens": 93, "prompt_tokens": 491, "total_tokens": 584, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 74, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014512, "usd": 0.0021447605194805197}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "9abce7868d0121fec78aee2ce78f90a3494ddffaf1eeec8b9334a4cfeb357637"}, "error": null}
|
||||
{"id": "train-de-005", "split": "train", "source_label": "de", "gold_label": "de", "text": "Der Hut kam sehr spät und fällt ziemlich klein aus. Er sieht aber gut aus und kann praktisch zusammen geklappt werden. Für den Preis - alles in Ordnung.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8771adf8be004bbe4b02", "response_model": "kimi-k3", "response_created": 1785366386, "request_started_at": "2026-07-29T23:06:24.845036+00:00", "latency_seconds": 4.247122874949127, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The user wants classification. Need output exactly one line Final Answer: de. We can reason privately: Text is German: Der Hut kam sehr spät... vocabulary Der, und, sehr, für den Preis, alles in Ordnung, umlaut ä. Label de. Ensure final exact format. No extra.", "prediction": "de", "usage": {"completion_tokens": 86, "prompt_tokens": 391, "total_tokens": 477, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 67, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011812, "usd": 0.0017457215584415586}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "6b947e8faea1fb8897f525e225e1d1db5f83593d6f8e8bf1e5732d76231bcb77"}, "error": null}
|
||||
{"id": "train-de-006", "split": "train", "source_label": "de", "gold_label": "de", "text": "Das Kleidchen ist wirklich schön. Allerdings wirklich nur was für den Strand. Ich trage normalerweise eine S. Hab mir eine M bestellt, damit es etwas lockerer sitzt, aber ich würde sagen, dass sogar eine L angebracht wäre.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8771f533e66c8281caac", "response_model": "kimi-k3", "response_created": 1785366386, "request_started_at": "2026-07-29T23:06:24.937664+00:00", "latency_seconds": 2.9810974579304457, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text is German: \"Das Kleidchen ist wirklich schön...\" vocabulary and function words clearly de. Final only line.", "prediction": "de", "usage": {"completion_tokens": 48, "prompt_tokens": 413, "total_tokens": 461, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 29, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008452, "usd": 0.0012491397402597402}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "6beb6733dd5f65d10c95796e921e09494971f1c868d4a74e139ff0d1f194b869"}, "error": null}
|
||||
{"id": "train-de-007", "split": "train", "source_label": "de", "gold_label": "de", "text": "Ich hatte diese Modell vor fünf Jahren gekauft. War damals super Qualität und hat auch lange gehalten. Jetzt das gleiche Modell von HAMA hier nachgekauft und große Enttäuschung. Kabel viel dünner als vor fünf Jahren. Verarbeitung viel schlechter. Nur günstige Materialien verwendet. Habe das Headset nun schon zum zweiten Mal umgetauscht und eine neues erhalten. Im täglichen Gebrauch hält es etwa zwei Wochen, dann schauen überall die Drähte raus und die Plastikteile lösen sich. Leider nicht mehr zu empfehlen !!!", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87713f3c9b5f06007edc", "response_model": "kimi-k3", "response_created": 1785366386, "request_started_at": "2026-07-29T23:06:25.022130+00:00", "latency_seconds": 2.838025875389576, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text is clearly German: Ich hatte, Qualität, gekauft, Enttäuschung, Verarbeitung, etc. Need final exact line.", "prediction": "de", "usage": {"completion_tokens": 53, "prompt_tokens": 503, "total_tokens": 556, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 34, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01536, "usd": 0.002270088311688312}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c81a7959c294dddbac74fb5f06fd33804da8bbcab83fb4234e8fc88a84b202cb"}, "error": null}
|
||||
{"id": "train-el-000", "split": "train", "source_label": "el", "gold_label": "el", "text": "ΠΊΝΑΚΑΣ 1 : σύγκριση των ηγετικών πρακτικών και των ομοσπονδιακών πρακτικών διαχείρισης των cio", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8772a5c09dd52de39314", "response_model": "kimi-k3", "response_created": 1785366386, "request_started_at": "2026-07-29T23:06:25.762588+00:00", "latency_seconds": 3.5992095000110567, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is clearly Greek: Greek script with words like ΠΊΝΑΚΑΣ (table), σύγκριση (comparison), των ηγετικών πρακτικών etc. \"cio\" is Latin acronym but negligible. Label el. Need final exact one line.", "prediction": "el", "usage": {"completion_tokens": 85, "prompt_tokens": 401, "total_tokens": 486, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 66, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011912, "usd": 0.0017605007792207796}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c823c8e70f97bb4b6cda41150148565d8641bf68a21efd4b48467d0c54b6c583"}, "error": null}
|
||||
{"id": "train-el-001", "split": "train", "source_label": "el", "gold_label": "el", "text": "Η ευελιξία του προϋπολογισμού μειώνεται δραστικά , έτσι ώστε μέχρι το 2050 , το καθαρό ενδιαφέρον για το χρέος θα απορροφήσει περίπου το ήμισυ όλων των ομοσπονδιακών εσόδων .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87733165d2c706d08be3", "response_model": "kimi-k3", "response_created": 1785366387, "request_started_at": "2026-07-29T23:06:26.438992+00:00", "latency_seconds": 2.257657625246793, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "Greek script.", "prediction": "el", "usage": {"completion_tokens": 24, "prompt_tokens": 443, "total_tokens": 467, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 5, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006652, "usd": 0.0009831137662337664}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c0a0c5d0683b843748559f138df8d19db47ef3228890c80e4697b5b992440fec"}, "error": null}
|
||||
{"id": "train-el-002", "split": "train", "source_label": "el", "gold_label": "el", "text": "Για όλο το ιστορικό μεγαλείο της , η πιστή γερουσία τώρα είναι το ισοδύναμο ενός Δημοτικού Συμβουλίου , η πολιτική του ικανότητα αφιερωμένη στις προμήθειες νερού , τις γραμμές αποχέτευσης και την ίδρυση παιδότοπους .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8773bfb3431a6ebc7f7d", "response_model": "kimi-k3", "response_created": 1785366387, "request_started_at": "2026-07-29T23:06:26.545520+00:00", "latency_seconds": 3.0944017497822642, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is in Greek script with Greek words: Για όλο το ιστορικό μεγαλείο, γερουσία, Δημοτικού Συμβουλίου etc. Label el.", "prediction": "el", "usage": {"completion_tokens": 72, "prompt_tokens": 466, "total_tokens": 538, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 53, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011912, "usd": 0.0017605007792207796}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "590b2a6abed1d1078d7285a2604ab52b5073f57581508e72d232b27991de93d6"}, "error": null}
|
||||
{"id": "train-el-003", "split": "train", "source_label": "el", "gold_label": "el", "text": "Θα πρέπει να πληρούνται οι απαιτήσεις για την κατάρτιση , την τεκμηρίωση και τη συντήρηση .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87743d093ac098615542", "response_model": "kimi-k3", "response_created": 1785366388, "request_started_at": "2026-07-29T23:06:27.560555+00:00", "latency_seconds": 2.9390022922307253, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text uses Greek script and Greek words: \"Θα πρέπει να πληρούνται οι απαιτήσεις...\" Clearly Greek. Label el. Need final exactly one line.", "prediction": "el", "usage": {"completion_tokens": 66, "prompt_tokens": 392, "total_tokens": 458, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 47, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009832, "usd": 0.0014530929870129872}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "60cba2331ab5a9f257444332f7288733003e9cdc559a124fa12399f2c4b5ed10"}, "error": null}
|
||||
{"id": "train-el-004", "split": "train", "source_label": "el", "gold_label": "el", "text": "Προστατευμένη από τους κρύο , υγρασία , βορειοδυτικά ανέμους από τα βουνά των vosges , οι αμπελώνες της αλσατία απολαμβάνουν ένα ιδανικό μικροκλίμα για την παραγωγή λευκών κρασιών που έχουν την εμπιστοσύνη τους στα πιο διάσημα κρασιά της βουργουνδίας και του μπορντώ .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877415c0ba78d0881f28", "response_model": "kimi-k3", "response_created": 1785366390, "request_started_at": "2026-07-29T23:06:27.569261+00:00", "latency_seconds": 5.367121750023216, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is in Greek script with Greek vocabulary: Προστατευμένη, τους, υγρασία, αμπελώνες, αλσατία, μικροκλίμα, λευκών κρασιών, βουργουνδίας, μπορντώ. Some capitalization/accent errors but clearly Greek. Label el. Need output exactly final line.", "prediction": "el", "usage": {"completion_tokens": 111, "prompt_tokens": 481, "total_tokens": 592, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 92, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016112, "usd": 0.002381228051948052}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "786abe8c185a65499cf4fc96ac67a7bd685b219729f1769725245aa44b0165b5"}, "error": null}
|
||||
{"id": "train-el-005", "split": "train", "source_label": "el", "gold_label": "el", "text": "Και κάναμε πολλή κηπουρική εκεί έξω και κυρίως σε μεγάλωσε κρεβάτια για να κρατήσουμε το χώμα ωραίο .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87754d654f2ee1729b34", "response_model": "kimi-k3", "response_created": 1785366389, "request_started_at": "2026-07-29T23:06:27.659914+00:00", "latency_seconds": 3.71215445920825, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is Greek script and words: \"Και κάναμε πολλή κηπουρική εκεί έξω...\" clearly Greek. Need final exactly one line. Ensure no extra.", "prediction": "el", "usage": {"completion_tokens": 66, "prompt_tokens": 403, "total_tokens": 469, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 47, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010052, "usd": 0.001485607272727273}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3bb9edf74a42be82283ab9cbc3daa878c8eea5b24586f6f06b13cb6aa947f92f"}, "error": null}
|
||||
{"id": "train-el-006", "split": "train", "source_label": "el", "gold_label": "el", "text": "Αχ μου αρέσει η γυναίκα μου δεν μπορεί να καταλάβει ότι θα είναι εκατό βαθμούς έξω θα είμαι εκεί έξω αλλά κάνει πολύ ζέστη για να δουλεύω στην αυλή είμαι κάτω από τα δέντρα περνάω καλά και σταματάω να πιω νερό οπότε τι είναι το πρόβλημα .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87740058eb9f85463804", "response_model": "kimi-k3", "response_created": 1785366389, "request_started_at": "2026-07-29T23:06:27.860919+00:00", "latency_seconds": 2.6522414167411625, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "Greek script, Greek vocabulary. Final Answer: el", "prediction": "el", "usage": {"completion_tokens": 31, "prompt_tokens": 476, "total_tokens": 507, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 12, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008012, "usd": 0.001184111168831169}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "aade171bc795eb9978987f8593b4ce0b52b98e70c359334140c802e706927ea4"}, "error": null}
|
||||
{"id": "train-el-007", "split": "train", "source_label": "el", "gold_label": "el", "text": "Απέναντι από το ο . Στο Audoen , δύο ενδιαφέροντες δρόμοι τρέχουν από την οδό Χάι .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8774b9c19ffa82a80147", "response_model": "kimi-k3", "response_created": 1785366389, "request_started_at": "2026-07-29T23:06:27.892059+00:00", "latency_seconds": 3.4703851668164134, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is Greek script with words like Απέναντι, από, το, Στο, δύο, ενδιαφέροντες, δρόμοι, τρέχουν, την, οδό. Clearly Greek. Need final exact line only.", "prediction": "el", "usage": {"completion_tokens": 84, "prompt_tokens": 393, "total_tokens": 477, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 65, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011652, "usd": 0.0017220748051948052}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "39beecab51639146658272cb64c334ec0e4a58e3dcd499ca780bade0f8d6b105"}, "error": null}
|
||||
{"id": "train-en-000", "split": "train", "source_label": "en", "gold_label": "en", "text": "This was ordered as a pack of 2 but I only received 1. Emailed the seller but never received a reply. Feeling jipped.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8775220baf5a77ff2236", "response_model": "kimi-k3", "response_created": 1785366390, "request_started_at": "2026-07-29T23:06:27.919452+00:00", "latency_seconds": 3.6247263746336102, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The text is English. Need final exact.", "prediction": "en", "usage": {"completion_tokens": 29, "prompt_tokens": 375, "total_tokens": 404, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 10, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005792, "usd": 0.0008560124675324677}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b30c2c0063e5e4e06533556f9eb4e3adf6f94ed380b2150478104e704fff7ca8"}, "error": null}
|
||||
{"id": "train-en-001", "split": "train", "source_label": "en", "gold_label": "en", "text": "Product received in damaged condition. Realized extent of damage after assembly almost completed since came in components. Unable to contact for product return.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8774157a4e90b05bc576", "response_model": "kimi-k3", "response_created": 1785366389, "request_started_at": "2026-07-29T23:06:28.231638+00:00", "latency_seconds": 4.1024928339757025, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The user wants language classification. The text is clearly English. Need final exactly one line Final Answer: en. Ensure no extra. Done. Now final.", "prediction": "en", "usage": {"completion_tokens": 51, "prompt_tokens": 371, "total_tokens": 422, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 32, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01252, "usd": 0.0018503584415584418}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "895a591132e0f6882bbe4e8a56bc29253a6c1e3d4c0c6a879e516092425d93f3"}, "error": null}
|
||||
{"id": "train-en-002", "split": "train", "source_label": "en", "gold_label": "en", "text": "Much cheaper quality than expected but item did arrive very fast and not damaged.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87755e7b32b0b9798f24", "response_model": "kimi-k3", "response_created": 1785366390, "request_started_at": "2026-07-29T23:06:28.697624+00:00", "latency_seconds": 2.7280842498876154, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The text is clearly English. Need final exactly one line Final Answer: en.", "prediction": "en", "usage": {"completion_tokens": 36, "prompt_tokens": 359, "total_tokens": 395, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 17, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006172, "usd": 0.0009121735064935067}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "d9091378a439888ea4db05588b60a2be975f84eee7b018c63887a9180686909f"}, "error": null}
|
||||
{"id": "train-en-003", "split": "train", "source_label": "en", "gold_label": "en", "text": "The book started out well and then just went on and on and on....... and never quite got to the point until I just quit reading it. I got to about 60% of the way and simply could not continue. I see where others have simply skipped ahead full chapters to get to the last 15% of the book. By the time I gave up my interest had simply disappeared. Too bad, could have been written in a much more concise and interesting format.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8775bcb8b1ca0dc259b5", "response_model": "kimi-k3", "response_created": 1785366390, "request_started_at": "2026-07-29T23:06:29.093002+00:00", "latency_seconds": 2.418180250097066, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The text is English, no ambiguity. Need final exactly one line format.", "prediction": "en", "usage": {"completion_tokens": 35, "prompt_tokens": 439, "total_tokens": 474, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 16, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007672, "usd": 0.0011338618181818182}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "171d25894318419f40fd01522cdc848feb4ce7e57eb695196685338f7c58b729"}, "error": null}
|
||||
{"id": "train-en-004", "split": "train", "source_label": "en", "gold_label": "en", "text": "Looked like it was going to be another homage to 80's horror in the vein of Ti West but it just doesn't know what it wants to be and wraps everything up in a really sloppy way. Setup, location and atmosphere are all there they just didn't know what to do with all of that after they jumped into the actual story. Kind of a disappointing miss since that is my only real complaint. All of the pieces were there for a great throwback horror movie and it was all squandered", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87750b8d682056dbef86", "response_model": "kimi-k3", "response_created": 1785366390, "request_started_at": "2026-07-29T23:06:29.362679+00:00", "latency_seconds": 3.08499920880422, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The user wants classification of the text. The text is clearly English: \"Looked like it was going to be another homage to 80's horror...\" Uses English vocabulary and grammar. Final answer must be exactly one line: Final Answer: en. No extra. Ensure format exact.", "prediction": "en", "usage": {"completion_tokens": 77, "prompt_tokens": 445, "total_tokens": 522, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 58, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011992, "usd": 0.001772324155844156}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "04e1efde7153560e12a327e935764b7bc6f148924afc96fc4fe1984c4cb00b9b"}, "error": null}
|
||||
{"id": "train-en-005", "split": "train", "source_label": "en", "gold_label": "en", "text": "Perfect for adding light layers of hydration especially in the colder months. This is usually my first layer of hydration before a moisturizer and it works great with no pilling or balling. Hydration lasts all day!", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877644ae93bbb7c864f2", "response_model": "kimi-k3", "response_created": 1785366391, "request_started_at": "2026-07-29T23:06:29.616015+00:00", "latency_seconds": 2.2913386248983443, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "English text.", "prediction": "en", "usage": {"completion_tokens": 23, "prompt_tokens": 386, "total_tokens": 409, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 4, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005412, "usd": 0.0007998514285714286}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "fe077218333fb4153fb8064e038e7e2b14c8df6bb50ab25280ab363d2fc46485"}, "error": null}
|
||||
{"id": "train-en-006", "split": "train", "source_label": "en", "gold_label": "en", "text": "Very big and gaudy looking. Will not wear them.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87766a87287a70e858ce", "response_model": "kimi-k3", "response_created": 1785366391, "request_started_at": "2026-07-29T23:06:29.640819+00:00", "latency_seconds": 2.5728488750755787, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "English.", "prediction": "en", "usage": {"completion_tokens": 22, "prompt_tokens": 357, "total_tokens": 379, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 3, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.004732, "usd": 0.0006993527272727274}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "de275da46549408372c8a6567ca871ab3f04946bfb1f6445fc8cf5061ff37d45"}, "error": null}
|
||||
{"id": "train-en-007", "split": "train", "source_label": "en", "gold_label": "en", "text": "Product description says gown but its an all in one pants and way top big for a newborn. The headband stitching is coming undone, very disappointed.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87760b8d682056dbef87", "response_model": "kimi-k3", "response_created": 1785366391, "request_started_at": "2026-07-29T23:06:30.500303+00:00", "latency_seconds": 1.9381071249954402, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The text is English. Need final only exact line. No extra.", "prediction": "en", "usage": {"completion_tokens": 34, "prompt_tokens": 374, "total_tokens": 408, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 15, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006272, "usd": 0.0009269527272727274}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "d19708f6a2b66312228ed2e2e156f939c9767cd309315bf964982e401d6848f3"}, "error": null}
|
||||
{"id": "train-es-000", "split": "train", "source_label": "es", "gold_label": "es", "text": "Por lo que vale y con premium te sacas la compra del mes. Trae funda y dos protectores por un gran precio.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8777577a2132f2c86b03", "response_model": "kimi-k3", "response_created": 1785366392, "request_started_at": "2026-07-29T23:06:30.513761+00:00", "latency_seconds": 3.4507440412417054, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "Spanish text: \"Por lo que vale... te sacas la compra del mes...\" -> es. Need final exact line.", "prediction": "es", "usage": {"completion_tokens": 46, "prompt_tokens": 374, "total_tokens": 420, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 27, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007472, "usd": 0.0011043033766233769}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "9ee54de7f88e6fdb372b458ca7fed8bef1ebcc8f98213fa2b60cf049b4ba02e3"}, "error": null}
|
||||
{"id": "train-es-001", "split": "train", "source_label": "es", "gold_label": "es", "text": "Son finas de algodón 100%, ideales para el veranito, yo las he comprado para regalar a una amiga que ha dado a luz hace un par de meses.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8778e3dfcfab6fd10197", "response_model": "kimi-k3", "response_created": 1785366393, "request_started_at": "2026-07-29T23:06:31.362736+00:00", "latency_seconds": 3.9875551252625883, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The text is Spanish: \"Son finas de algodón 100%, ideales para el veranito, yo las he comprado para regalar a una amiga que ha dado a luz hace un par de meses.\" Clear Spanish vocabulary and grammar. Output label es.", "prediction": "es", "usage": {"completion_tokens": 81, "prompt_tokens": 389, "total_tokens": 470, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 62, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011272, "usd": 0.0016659137662337665}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "d940f3683eea7e00e545081eedb5605946689b58757c924b4f9cf610617d2de1"}, "error": null}
|
||||
{"id": "train-es-002", "split": "train", "source_label": "es", "gold_label": "es", "text": "Buenísimos. Pintan genial. Los he probado sobre cristal y tienen una solidez que permite que se vean perfectamente. Se mantienen durante mucho tiempo pero a la vez de borran con facilidad. Los recomiendo 100%. De hecho será la marca que compre a partir de ahora.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87777d03af917e2edc21", "response_model": "kimi-k3", "response_created": 1785366392, "request_started_at": "2026-07-29T23:06:31.372373+00:00", "latency_seconds": 3.453065207693726, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The text is in Spanish. Words like \"Buenísimos\", \"Pintan genial\", \"Los he probado\", \"cristal\", \"solidez\", \"Se mantienen\", \"borran con facilidad\", \"Los recomiendo\", \"marca que compre\" are clearly Spanish vocabulary and grammar. Final answer es.", "prediction": "es", "usage": {"completion_tokens": 92, "prompt_tokens": 419, "total_tokens": 511, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 73, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012972, "usd": 0.0019171605194805196}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b19ef08ce346e1982b568fa5a73657905dcf478796b3394ff2df953e02ef8df8"}, "error": null}
|
||||
{"id": "train-es-003", "split": "train", "source_label": "es", "gold_label": "es", "text": "Me ha gustado mucho tanto la calidad de los materiales como los acabados que tiene, el toque de la madera es muy chulo. Tiene las clásicas tres velocidades aunque como ya sabéis se usa en el 90% de las veces la misma. Se puede colocar unos altavoces por la parte de atrás, la verdad que suena bastante más alto de lo que me pensaba, aún así he conectado los dos altavoces que tengo repartidos por el comedor y el resultado es espectacular. Unos recuerdos del sonido clásico que me ha gustado mucho recordar, estoy muy contento con el resultado. A destacar la superficie de la maleta, estéticamente es muy bonita, incluso simplemente para tener colocado encima de algún mueble, cumple perfectamente los dos cometidos, la estética y la funcional. Por supuesto cuenta con entrada USB y con función bluetooth, perfecto porque puedes reproducir también la música del teléfono móvil.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8778dac2ed2f421575cc", "response_model": "kimi-k3", "response_created": 1785366392, "request_started_at": "2026-07-29T23:06:31.426384+00:00", "latency_seconds": 2.91424870910123, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The text is clearly Spanish: \"Me ha gustado mucho\", \"la calidad\", \"los materiales\", \"está\", \"muy\", etc. Spanish vocabulary and grammar throughout. Final Answer: es.", "prediction": "es", "usage": {"completion_tokens": 65, "prompt_tokens": 594, "total_tokens": 659, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 46, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013772, "usd": 0.002035394285714286}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "4c5b82b0966c668f78f8599303545bbeb1dd527bdfaa6601ab1cfe0f519918c2"}, "error": null}
|
||||
{"id": "train-es-004", "split": "train", "source_label": "es", "gold_label": "es", "text": "Muy robusto, era para niño de 8 años pero como es regulable le servirá para mucho", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8778db7d851280a77052", "response_model": "kimi-k3", "response_created": 1785366393, "request_started_at": "2026-07-29T23:06:31.512029+00:00", "latency_seconds": 3.5647654589265585, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The text is Spanish: \"Muy robusto, era para niño de 8 años pero como es regulable le servirá para mucho\". Clear Spanish vocabulary and function words. Answer es.", "prediction": "es", "usage": {"completion_tokens": 62, "prompt_tokens": 370, "total_tokens": 432, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 43, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008992, "usd": 0.0013289475324675325}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3480bc0744aab2b15d929f2a41bcbd161925f934c1bcfae91366d8b434d71f7e"}, "error": null}
|
||||
{"id": "train-es-005", "split": "train", "source_label": "es", "gold_label": "es", "text": "Me gusta brillo de labios, es muy poca cantidad pero esta bien.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8778aff9f7f7f84cf43b", "response_model": "kimi-k3", "response_created": 1785366393, "request_started_at": "2026-07-29T23:06:31.544934+00:00", "latency_seconds": 3.2186344591900706, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "Spanish: \"Me gusta... es muy poca cantidad pero esta bien.\" Label es. Ensure final only.", "prediction": "es", "usage": {"completion_tokens": 44, "prompt_tokens": 363, "total_tokens": 407, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 25, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007052, "usd": 0.0010422306493506495}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "87d6c11804ee67b7105c28d9c6b862ce289967ed82f75a2479cfa144ec380dec"}, "error": null}
|
||||
{"id": "train-es-006", "split": "train", "source_label": "es", "gold_label": "es", "text": "No es compatible para el nuevo XS, todas las fundas son del X y las venden como XS y no saben que no es igual, solo hay que ver el altavoz de al rededor de la cámara", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87784d654f2ee1729b38", "response_model": "kimi-k3", "response_created": 1785366393, "request_started_at": "2026-07-29T23:06:31.907982+00:00", "latency_seconds": 3.2280592499300838, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "Text is clearly Spanish: \"No es compatible para el nuevo XS...\" Function words es, para, el, las, son, del, y, que, no, igual. Label es. Need output exactly final answer line. Ensure no extra.", "prediction": "es", "usage": {"completion_tokens": 70, "prompt_tokens": 392, "total_tokens": 462, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 51, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010232, "usd": 0.0015122098701298703}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0298f3f7116773114739ceb2e3df10b321bd0fcae6ebcddfccf7a4cb90f51941"}, "error": null}
|
||||
{"id": "train-es-007", "split": "train", "source_label": "es", "gold_label": "es", "text": "Esta bien pero pense que incluía las bombillas...", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8779124dbf4a23c9891a", "response_model": "kimi-k3", "response_created": 1785366394, "request_started_at": "2026-07-29T23:06:32.214316+00:00", "latency_seconds": 4.127376291900873, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The user wants classify. Text: \"Esta bien pero pense que incluía las bombillas...\" Clearly Spanish: esta bien, pero, pensé que incluía las bombillas (bulbs). Missing accents but Spanish. Need output exactly Final Answer: es. Done.", "prediction": "es", "usage": {"completion_tokens": 78, "prompt_tokens": 358, "total_tokens": 436, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 59, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010352, "usd": 0.0015299449350649353}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "d6c9ff9e76b3cdfb11abcc4a9402bf103fe7fcd6d9f43f3ab0519a8aebab14a3"}, "error": null}
|
||||
{"id": "train-fr-000", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Bien reçu bien emballer fonctionne correctement, j’avais peur du bruit par rapport au commentaires mais le bruit des ventilateurs est correct", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87790a128ec1a6782ce7", "response_model": "kimi-k3", "response_created": 1785366394, "request_started_at": "2026-07-29T23:06:32.334828+00:00", "latency_seconds": 3.918092292267829, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The user wants classification. Text is French: \"Bien reçu bien emballer fonctionne correctement, j’avais peur du bruit...\" Clear French vocabulary and grammar. Need output exactly Final Answer: fr. Ensure one line.", "prediction": "fr", "usage": {"completion_tokens": 72, "prompt_tokens": 383, "total_tokens": 455, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 53, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010252, "usd": 0.0015151657142857145}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3b3696d888ed950f4ef7fec8979f31ba5d1aa4c11ed3e850da07dadd4582c1ba"}, "error": null}
|
||||
{"id": "train-fr-001", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Ce produit est tout simplement génial. il agit très vite. l'odeur disparait quasiment au moment de la pulvérisation, il n'y a pas pas besoin d'en mettre beaucoup le flacon dure longtemps. Ensuite le nettoyage est très facile et il y a même comme une petite odeur de frais. Le chat ne reviens pas faire au même endroit après avoir nettoyé avec ce produit. N'hésitez pas.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877aa22fab7265b5be39", "response_model": "kimi-k3", "response_created": 1785366395, "request_started_at": "2026-07-29T23:06:32.439519+00:00", "latency_seconds": 3.795263625215739, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text is French. Need output exactly one line Final Answer: fr. Ensure no extra.", "prediction": "fr", "usage": {"completion_tokens": 39, "prompt_tokens": 456, "total_tokens": 495, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 20, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008412, "usd": 0.001243228051948052}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "097fded44feb926335c172cd00831481fc85209b12edf03938d646b14e86cd9e"}, "error": null}
|
||||
{"id": "train-fr-002", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Beau mais pour coller c'est pas très efficace. J'en ai essayé 2 ils n'ont pas tenu 1 semaine. Mais j'en ai fixé un avec une vis. Donc on peut l'adapter.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877912e92b55023aa874", "response_model": "kimi-k3", "response_created": 1785366394, "request_started_at": "2026-07-29T23:06:32.448244+00:00", "latency_seconds": 4.583942916244268, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text is clearly French: \"Beau mais pour coller c'est pas très efficace. J'en ai essayé 2 ils n'ont pas tenu 1 semaine. Mais j'en ai fixé un avec une vis. Donc on peut l'adapter.\" French function words, accents é, ç, c'est, j'en. Final answer fr.", "prediction": "fr", "usage": {"completion_tokens": 98, "prompt_tokens": 396, "total_tokens": 494, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 79, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013112, "usd": 0.0019378514285714288}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "cf70db6b935c47da13e2114c5e4981f4d4bc611644ec292f438abd9e99dc7035"}, "error": null}
|
||||
{"id": "train-fr-003", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "L' idée m' avait séduit mais l' objet est décevant. Les chiffres sont de simples autocollants , la rotation des pièces n' est pas fluide. Je n' aurais pas acheté si j' avais pu le toucher.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87793165d2c706d08bee", "response_model": "kimi-k3", "response_created": 1785366394, "request_started_at": "2026-07-29T23:06:32.937379+00:00", "latency_seconds": 3.25823837518692, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text is clearly French: \"L'idée m'avait séduit mais l'objet est décevant...\" Contains French vocabulary, accents, function words. Label fr. Need final exactly one line. Ensure no extra.", "prediction": "fr", "usage": {"completion_tokens": 67, "prompt_tokens": 407, "total_tokens": 474, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 48, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010232, "usd": 0.0015122098701298703}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "cae23e0b9af18058686bf18322a50c734948b71d8dbcfd3d578b27efb58ec9a8"}, "error": null}
|
||||
{"id": "train-fr-004", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Torche puissante, idéale pour un vélo, le système de fixation semble fiable. En revanche, je regrette la batterie fournie et nécessaire qui ne correspond à aucune classe de pile classique.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877b98f2260e94fe9ad9", "response_model": "kimi-k3", "response_created": 1785366397, "request_started_at": "2026-07-29T23:06:33.965138+00:00", "latency_seconds": 4.628089042380452, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text is French: 'Torche puissante, idéale pour un vélo...' Vocabulary and diacritics clearly French. Need final exactly one line.", "prediction": "fr", "usage": {"completion_tokens": 57, "prompt_tokens": 396, "total_tokens": 453, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 38, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009012, "usd": 0.0013319033766233767}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "77f9853a8882fc9fd1f6459c7d6581488a3321a63b6008fcda1c88daffe9c5cd"}, "error": null}
|
||||
{"id": "train-fr-005", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "LIVRAISON RAPIDE ET SERIEUSE", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877bd657b69fbc2d80a5", "response_model": "kimi-k3", "response_created": 1785366396, "request_started_at": "2026-07-29T23:06:34.341286+00:00", "latency_seconds": 3.684275375213474, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text \"LIVRAISON RAPIDE ET SERIEUSE\" is French: livraison = delivery, rapide = fast, et = and, sérieuse = serious/reliable. Clear French vocabulary and function word \"et\". Label fr.", "prediction": "fr", "usage": {"completion_tokens": 72, "prompt_tokens": 356, "total_tokens": 428, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 53, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009712, "usd": 0.0014353579220779222}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "2355b9d7f3cb045bf22f8ace12acc394754fa195d75635c3f75150f70ffa9f70"}, "error": null}
|
||||
{"id": "train-fr-006", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Je viens de m'apercevoir que ce produit vendu comme neuf semble être reconditionné puisque une photo est restée dans l'appareil (personnes et lieu que je ne connais pas du tout) . La date de la photo est antérieure à ma date d'achat.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877b15c0ba78d0881f2d", "response_model": "kimi-k3", "response_created": 1785366395, "request_started_at": "2026-07-29T23:06:34.764362+00:00", "latency_seconds": 1.8489160840399563, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "French clearly.", "prediction": "fr", "usage": {"completion_tokens": 23, "prompt_tokens": 414, "total_tokens": 437, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 4, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005972, "usd": 0.000882615064935065}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "edecdcd9c7f733911d6eae68e8f92ec0348073a14e6206c615ac690e6f8bbc2d"}, "error": null}
|
||||
{"id": "train-fr-007", "split": "train", "source_label": "fr", "gold_label": "fr", "text": "Simple jouet avec lequel on s'amuse deux minutes avec , la mode est finis", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877b419b7910c2fd95e0", "response_model": "kimi-k3", "response_created": 1785366396, "request_started_at": "2026-07-29T23:06:34.826459+00:00", "latency_seconds": 4.833099541720003, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The user wants me to classify the language of the given text. The text is: \"Simple jouet avec lequel on s'amuse deux minutes avec , la mode est finis\"\n\nThis contains French words: \"jouet\" (toy), \"avec\" (with), \"lequel\" (which), \"on s'amuse\" (we have fun), \"deux minutes\" (two minutes), \"la mode est finis\" (the fashion is over). This is clearly French, despite some grammatical errors.\n\nThe answer should be fr.", "prediction": "fr", "usage": {"completion_tokens": 134, "prompt_tokens": 364, "total_tokens": 498, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 115, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016072, "usd": 0.002375316363636364}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "07a781503616a7fdad80c3373c7100d881775e0cb62963cc8fdf535c4badde76"}, "error": null}
|
||||
{"id": "train-hi-000", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "हम प ् रत ् येक राज ् य के लिए कर कोड और प ् रक ् रियाओं को प ् राप ् त कर सकते हैं , इन ् हें जांच करें , चुनिंदा अधिकारियों का साक ् षात ् कार करें और कुछ विश ् वसनीय पैटर ् न उत ् पन ् न करें .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877ba5c09dd52de3931d", "response_model": "kimi-k3", "response_created": 1785366396, "request_started_at": "2026-07-29T23:06:35.077539+00:00", "latency_seconds": 3.430446374695748, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script. The words are Hindi: हम प्रत्येक राज्य के लिए कर कोड और प्रक्रियाओं को प्राप्त कर सकते हैं... This is clearly Hindi despite spacing artifacts around conjuncts. Label hi. Need final exactly one line.", "prediction": "hi", "usage": {"completion_tokens": 87, "prompt_tokens": 459, "total_tokens": 546, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 68, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013272, "usd": 0.0019614981818181824}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "edeb505dfe58512278ed1d28136c1c7b0d05a60fc2a87a7809ed2eb87dd5f600"}, "error": null}
|
||||
{"id": "train-hi-001", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "और उम हाँ उह हाँ मुझे लगता है कि मैं आमतौर पर मैं हूँ उम भारी सॉस और उम", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877ce5cd1f53d10ec52e", "response_model": "kimi-k3", "response_created": 1785366396, "request_started_at": "2026-07-29T23:06:35.136902+00:00", "latency_seconds": 4.189001750200987, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text uses Devanagari script with Hindi words: और, हाँ, मुझे लगता है, मैं, आमतौर पर, भारी सॉस. Disfluencies उम/उह are Hindi filler. Clearly Hindi. Final line only.", "prediction": "hi", "usage": {"completion_tokens": 82, "prompt_tokens": 391, "total_tokens": 473, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 63, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011412, "usd": 0.0016866046753246755}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "625440b74758082f49cf4689c90135bbf549b2e7c92fd9a3a94373635dc10d5c"}, "error": null}
|
||||
{"id": "train-hi-002", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "जून में , जो कि जून में अपेक ् षित है , वह व ् हाइट हाउस वकील के जनादेश के न ् यायालय की व ् याख ् या को चालू करेगा . इस निर ् णय का कोई कानूनी उदाहरण नहीं है .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877cbe759e5d2a00e61e", "response_model": "kimi-k3", "response_created": 1785366397, "request_started_at": "2026-07-29T23:06:35.350974+00:00", "latency_seconds": 4.5833537499420345, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script, Hindi vocabulary: जून, जो कि, अपेक्षित, व्हाइट हाउस वकील, जनादेश, न्यायालय, व्याख्या, निर्णय, कानूनी उदाहरण. Clearly Hindi despite spacing artifacts. Need final exact format.", "prediction": "hi", "usage": {"completion_tokens": 100, "prompt_tokens": 437, "total_tokens": 537, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 81, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014132, "usd": 0.002088599480519481}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "112bef9137cbcd372ab3e7eb8c01436afa0f752b701d706c5d820a357bba83c5"}, "error": null}
|
||||
{"id": "train-hi-003", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "Anao की साइट में उनके विभिन ् न प ् रकाशनों के लिंक शामिल हैं , जिसमें अपनी लेखापरीक ् षा रिपोर ् ट और बेहतर अभ ् यास गाइड शामिल हैं .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877cde3406f5dc3ef76d", "response_model": "kimi-k3", "response_created": 1785366397, "request_started_at": "2026-07-29T23:06:36.196285+00:00", "latency_seconds": 4.378484250046313, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text contains Devanagari script with Hindi words: की साइट में उनके विभिन्न प्रकाशनों के लिंक शामिल हैं, जिसमें अपनी लेखापरीक्षा रिपोर्ट और बेहतर अभ्यास गाइड शामिल हैं. Starts with \"Anao\" (proper noun/URL-ish) but rest is Hindi. Classify hi.", "prediction": "hi", "usage": {"completion_tokens": 115, "prompt_tokens": 421, "total_tokens": 536, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 96, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015312, "usd": 0.002262994285714286}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7065f4dcdea2cee1775edc25e92127e7e39c424e1077b68b93e37ee52218e9bc"}, "error": null}
|
||||
{"id": "train-hi-004", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "11 वीं राजवंश की अवधि से पीटर के प ् राचीन मिस ् री धर ् म , सी . 2134 ईसा पूर ् व", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877e628dc39d96c21c28", "response_model": "kimi-k3", "response_created": 1785366398, "request_started_at": "2026-07-29T23:06:36.235667+00:00", "latency_seconds": 4.279816957656294, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text uses Devanagari script with Hindi words: \"11 वीं राजवंश की अवधि\" etc. Clear Hindi. Need output exact format. Ensure no extra reasoning in final.", "prediction": "hi", "usage": {"completion_tokens": 63, "prompt_tokens": 397, "total_tokens": 460, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 44, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009632, "usd": 0.0014235345454545456}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "63394ccffb42b56ffc7164ba20bd74c482f901d00ef8f705796bb9a802d768b0"}, "error": null}
|
||||
{"id": "train-hi-005", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "हाँ हाँ किसी ने कहा कि उम कि इस तरह के कुछ प ् रकार के शुरू में कुछ प ् रकार की है और यह कि जब कार ठंडा हो जाता है , तो धातु को छोटा बनाने के लिए फैलता है और तब तक यह नहीं है कि कार को नीचे ठंडा होने तक नहीं है धातु आप जानते हैं कि इन दो टुकड ़ े किसी और को छू नहीं रहे हैं और आप और आप जानते हैं कि आप जानते हैं तो यह शुरू हो जाएगा क ् योंकि आपके पास छोटा नहीं है", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877d4cb5f60e0c84468e", "response_model": "kimi-k3", "response_created": 1785366398, "request_started_at": "2026-07-29T23:06:36.253628+00:00", "latency_seconds": 6.607215458061546, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script. Need distinguish Hindi vs Urdu. It uses Devanagari, common Hindi words: हाँ, किसी ने कहा, इस तरह, शुरू, कार ठंडा, धातु, फैलता, etc. Some spacing artifacts around conjuncts (प ् रकार) but clearly Hindi. Not Urdu (usually Arabic script), though spoken content could be Hindi/Urdu; native script Devanagari -> hi. Need final exact line only.", "prediction": "hi", "usage": {"completion_tokens": 133, "prompt_tokens": 546, "total_tokens": 679, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 114, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.019612, "usd": 0.0028985007792207795}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "512e884967f46b6f25503dd2efdae4cb627797f6f22446bf35db69cc9c914369"}, "error": null}
|
||||
{"id": "train-hi-006", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "लेखा विषयों के अतिरिक ् त , निर ् देशिका सूची में कुछ ऐसी एजेंसियों या प ् रोग ् राम जो उदाहरण में प ् रयोग किया गया है या उसके मानकों के भीतर अद ् वितीय प ् रावधान है .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877dbcc40607e9f8ab29", "response_model": "kimi-k3", "response_created": 1785366398, "request_started_at": "2026-07-29T23:06:36.342333+00:00", "latency_seconds": 4.4440536657348275, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script with Hindi vocabulary and grammar: \"लेखा विषयों के अतिरिक्त\", \"निर्देशिका सूची\", \"उदाहरण\", \"प्रयोग किया गया है\", \"मानकों\", \"अद्वितीय प्रावधान\". Clearly Hindi. Final answer must be exactly one line.", "prediction": "hi", "usage": {"completion_tokens": 106, "prompt_tokens": 442, "total_tokens": 548, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 87, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014832, "usd": 0.002192054025974026}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "9bf1b27ccd6bc2f71b7fdb87a99cff47bb7825385b89af9e1624b20aeaa1732a"}, "error": null}
|
||||
{"id": "train-hi-007", "split": "train", "source_label": "hi", "gold_label": "hi", "text": "वे चीज ़ ें जो केवल आप को प ् रभावित कर रहे हैं , उम कि हम लोगों को कैसे दंडित करते हैं और क ् यों चीजें हैं जैसे कि वे रास ् ते हैं", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877e628dc39d96c21c2a", "response_model": "kimi-k3", "response_created": 1785366399, "request_started_at": "2026-07-29T23:06:36.613606+00:00", "latency_seconds": 6.5006488328799605, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script with Hindi words like वे, चीजें, जो, केवल, आप, को, प्रभावित, कर, रहे, हैं, उम, कि, हम, लोगों, को, कैसे, दंडित, करते, हैं, और, क्यों, चीजें, हैं, जैसे, कि, वे, रास्ते, हैं. It has some spacing artifacts around combining marks but clearly Hindi. Final label hi.", "prediction": "hi", "usage": {"completion_tokens": 139, "prompt_tokens": 420, "total_tokens": 559, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 120, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.017692, "usd": 0.0026147397402597403}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "1ea512166c75a6d8b89e12e2808cf3ccb5e3706eb996ed452a28d9c1c8814ba0"}, "error": null}
|
||||
{"id": "train-it-000", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Gli aerei da guerra russi colpiscono in Siria", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877e4cff86af330ad142", "response_model": "kimi-k3", "response_created": 1785366399, "request_started_at": "2026-07-29T23:06:37.033012+00:00", "latency_seconds": 3.740647125057876, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Italian: \"Gli aerei da guerra russi colpiscono in Siria\" meaning \"Russian warplanes strike in Syria\". Italian is not in the listed labels, so classify as ot. Need final exact line only.", "prediction": "ot", "usage": {"completion_tokens": 71, "prompt_tokens": 361, "total_tokens": 432, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 52, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009712, "usd": 0.0014353579220779222}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "36da1f9ea097c21abede6d7c0d29f9471838c1c679f9fae71722f72360a382df"}, "error": null}
|
||||
{"id": "train-it-001", "split": "train", "source_label": "it", "gold_label": "ot", "text": "L'uomo sta portando una cassetta degli attrezzi sul marciapiede.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877eccc0dbe1b98e03e2", "response_model": "kimi-k3", "response_created": 1785366399, "request_started_at": "2026-07-29T23:06:38.026524+00:00", "latency_seconds": 2.8794856248423457, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Italian: \"L'uomo sta portando una cassetta degli attrezzi sul marciapiede.\" Italian is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 67, "prompt_tokens": 366, "total_tokens": 433, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 48, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009412, "usd": 0.0013910202597402599}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "78486a6f6bab5be2495030ba69b05a5878294086c6b7ce0e32e940db10a627cb"}, "error": null}
|
||||
{"id": "train-it-002", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Israele trattiene 37 palestinesi mentre prosegue l'operazione di arresto", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877f9904a4c085eca381", "response_model": "kimi-k3", "response_created": 1785366400, "request_started_at": "2026-07-29T23:06:38.508700+00:00", "latency_seconds": 4.311743167228997, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Italian: \"Israele trattiene 37 palestinesi mentre prosegue l'operazione di arresto\" - clearly Italian vocabulary (trattiene, mentre, prosegue, operazione). Italian is not in the listed labels, so classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 81, "prompt_tokens": 368, "total_tokens": 449, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 62, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010852, "usd": 0.0016038410389610392}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "30d7365ff0fad5134b2465c7e46ee6037a5bf2cebf32ab6c92c0f62e2b05766f"}, "error": null}
|
||||
{"id": "train-it-003", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Uomo ucciso in un raid del terrore francese", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a877f80beffa1ad578f66", "response_model": "kimi-k3", "response_created": 1785366400, "request_started_at": "2026-07-29T23:06:38.593857+00:00", "latency_seconds": 3.1493472917936742, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"Uomo ucciso in un raid del terrore francese\" is Italian. Italian is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 60, "prompt_tokens": 359, "total_tokens": 419, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 41, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008572, "usd": 0.0012668748051948052}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8bcb54cdc3e361a089bd0b3ee64308800ca22316dfdb190000ebb6ffe3ac8600"}, "error": null}
|
||||
{"id": "train-it-004", "split": "train", "source_label": "it", "gold_label": "ot", "text": "La gente va e pagaia su una zattera.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87805657fe29c59a3a5c", "response_model": "kimi-k3", "response_created": 1785366400, "request_started_at": "2026-07-29T23:06:39.326673+00:00", "latency_seconds": 2.805106417275965, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Italian: \"La gente va e pagaia su una zattera.\" Not in list → ot.", "prediction": "ot", "usage": {"completion_tokens": 43, "prompt_tokens": 358, "total_tokens": 401, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 24, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006852, "usd": 0.0010126722077922078}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e31162ad6b82b7f708cb363225c6822068a0789b3ced6e6f9b7e8b49142c4ebd"}, "error": null}
|
||||
{"id": "train-it-005", "split": "train", "source_label": "it", "gold_label": "ot", "text": "Mi oppongo alla pena di morte.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87819eab2b3298a1c572", "response_model": "kimi-k3", "response_created": 1785366401, "request_started_at": "2026-07-29T23:06:39.660199+00:00", "latency_seconds": 2.7062623747624457, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Italian not in list => ot.", "prediction": "ot", "usage": {"completion_tokens": 27, "prompt_tokens": 354, "total_tokens": 381, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 8, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005172, "usd": 0.0007643812987012988}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "646448c11aacd436a7886bb7fa6f0b31eadea3e8d0a2237e01ce862d4422c0b7"}, "error": null}
|
||||
{"id": "train-it-006", "split": "train", "source_label": "it", "gold_label": "ot", "text": "La donna sta condendo l'olio.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8780567e9c2404b6c4c6", "response_model": "kimi-k3", "response_created": 1785366402, "request_started_at": "2026-07-29T23:06:39.935346+00:00", "latency_seconds": 6.482084542047232, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"La donna sta condendo l'olio.\" is Italian: \"The woman is seasoning the oil\" / \"La donna\" (the woman), \"sta condendo\" (is seasoning/dressing), \"l'olio\" (the oil). Italian is not in the listed labels, so classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 88, "prompt_tokens": 354, "total_tokens": 442, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 69, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011272, "usd": 0.0016659137662337665}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "be79d5528a1eaf35958d2525fd37b25e0a020c43ef9cf0fccb9f02bc26e1dc92"}, "error": null}
|
||||
{"id": "train-it-007", "split": "train", "source_label": "it", "gold_label": "ot", "text": "In una dichiarazione via e-mail al Knoxville News Sentinel, Shumaker ha detto: \"Non prendo in considerazione le dimissioni.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87813165d2c706d08c01", "response_model": "kimi-k3", "response_created": 1785366401, "request_started_at": "2026-07-29T23:06:40.516265+00:00", "latency_seconds": 2.015511500183493, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Italian, not in list -> ot.", "prediction": "ot", "usage": {"completion_tokens": 28, "prompt_tokens": 376, "total_tokens": 404, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 9, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005712, "usd": 0.000844189090909091}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a6fef96c260979e1df01c8f4f6658384e0d573b45e09bbe11210e5558482e1a9"}, "error": null}
|
||||
{"id": "train-ja-000", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "味はとても良かったのですが、もうちょっと手頃な価格だと良かったのですが・・・", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87815a513bb219e71373", "response_model": "kimi-k3", "response_created": 1785366401, "request_started_at": "2026-07-29T23:06:40.575508+00:00", "latency_seconds": 4.4139669998548925, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user wants me to classify text. The text is Japanese: \"味はとても良かったのですが、もうちょっと手頃な価格だと良かったのですが・・・\" Japanese is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh), so it's \"ot\". Final answer must be exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 114, "prompt_tokens": 379, "total_tokens": 493, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 95, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014372, "usd": 0.0021240696103896105}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5f68bb907aa2d88fc8ec23e1e7b1a3229bb835b78449928d436e1e2dce604fc3"}, "error": null}
|
||||
{"id": "train-ja-001", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "ジュースに入れて使用しています。まだ加熱料理にしようしていませんが、機会があれば使用してみたいと思います。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878233bb1483514480b9", "response_model": "kimi-k3", "response_created": 1785366402, "request_started_at": "2026-07-29T23:06:40.774367+00:00", "latency_seconds": 4.836877999827266, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese: \"ジュースに入れて使用しています。まだ加熱料理にしようしていませんが、機会があれば使用してみたいと思います。\" Japanese is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh), so it falls under ot (all other languages). Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 112, "prompt_tokens": 382, "total_tokens": 494, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 93, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014232, "usd": 0.0021033787012987013}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a0354cd999cfcb1b2f62b857e658330daf20d78f5773d1ed40958cbee32a23a9"}, "error": null}
|
||||
{"id": "train-ja-002", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "７キロ用の洗濯機にはパツンパツンです。表記のサイズはいい加減なので、１〜２cm小さいと考えてください。シャープのプラズマクラスター付きの現行品にテラスで使用のため、このカバーを買いましたが強風で上方にズレるため、引っ張って下に降ろす際に、すでに裏の結び紐の片側が切れてしまいました。簡単に切れるので気を付けたいですね？！ なので価格のことを考えると、６ヶ月（新年を迎えられれば）御の字ですが、果たして３ヶ月持つかどうか？様子見です。前のものがLEC（レック）という老舗のブランド（上場企業なのにアフターの対応も最悪なので２度とこのメーカーの製品は使わないと決めています）のもので１２００円位した割りには、１５ヶ月位しか持たなかったので、３〜４ヶ月使えればと。 →結果として１ヶ月でボロボロ。既に廃棄しました。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87814cb5f60e0c844698", "response_model": "kimi-k3", "response_created": 1785366402, "request_started_at": "2026-07-29T23:06:40.787023+00:00", "latency_seconds": 3.946453125216067, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese (hiragana, katakana, kanji). Japanese is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh). Chinese is zh but this is Japanese, not Chinese. So label ot.", "prediction": "ot", "usage": {"completion_tokens": 83, "prompt_tokens": 658, "total_tokens": 741, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 64, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016852, "usd": 0.0024905942857142856}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7f34facb5e567ad9137302705ee5c14c17a648f8cabcc05845d43df393b128b2"}, "error": null}
|
||||
{"id": "train-ja-003", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "一月過ぎたけどまだ来ない。最悪‼️流石中国星1つ付けたけど本当は星-5", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8781a95a610865f1e377", "response_model": "kimi-k3", "response_created": 1785366402, "request_started_at": "2026-07-29T23:06:40.906775+00:00", "latency_seconds": 5.514311916194856, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese: \"一月過ぎたけどまだ来ない。最悪‼️流石中国星1つ付けたけど本当は星-5\". It uses hiragana/katakana/kanji, Japanese grammar (過ぎたけど, 来ない, 最悪, 流石). Contains word 中国 (China) but language is Japanese. Japanese is not in listed labels, so ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 125, "prompt_tokens": 380, "total_tokens": 505, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 106, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015492, "usd": 0.0022895968831168835}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "89ec4dadeb2d38197e570d587a8ca9791b2b16a946e1e9def459b730e8b6e948"}, "error": null}
|
||||
{"id": "train-ja-004", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "埃に気を付け、湿度のある浴室で貼り付けを行いました。埃は問題が無かったのですが、残念ながら他の方も書かれていますが端の方に気泡が残ります、ガラス面と黒い額縁の境目に残ります。 爪で押し出しましたが気泡を追い出せません。 少し剥がし、気泡を追い出しましたがそういう問題で気泡が残っているわけでは無いようで、若干浮いていて気泡が入っている様です。 穴の部分はぴったりの位置にあるのでRの部分が相当シビアに作られているのかも知れません。 うまく貼れた場合でも、使用するとガラス面に結構指紋が付きます、何も貼らなかった場合に比べて滑らかさと防汚の面が劣りますが、これは全てのこういった商品に当てはまるのかも知れません。 ガラスの傷つきにくさですが、何も貼らなかった場合に比べて貼った方が強度は上のようです。 1ヵ月、XZ1をそのまま使用していると光に当てないと分からない位ですが、小傷が沢山出来ていました、こちらの商品を貼って1週間色々ハードに使いましたが傷は出来ておりません、強度は申し分ありません。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87824d654f2ee1729b41", "response_model": "kimi-k3", "response_created": 1785366402, "request_started_at": "2026-07-29T23:06:41.743869+00:00", "latency_seconds": 2.606724292039871, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Japanese (hiragana/katakana/kanji), which is not in the listed labels, so classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 49, "prompt_tokens": 710, "total_tokens": 759, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 30, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014492, "usd": 0.0021418046753246757}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a0af90c8d04c4bfbc7e5ce290a807e02c080825233a5cb9e7952ae28da95ef62"}, "error": null}
|
||||
{"id": "train-ja-005", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "マツエク休憩中で数年ぶりにマスカラを買いました。ダマにならず、しっかり付くのに本当にお湯でスルッと取れるので良いです！", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878215c0ba78d0881f34", "response_model": "kimi-k3", "response_created": 1785366403, "request_started_at": "2026-07-29T23:06:42.132550+00:00", "latency_seconds": 2.4985699579119682, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese (hiragana/katakana/kanji). Japanese is not in the listed labels, so classify as ot. Final answer exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 54, "prompt_tokens": 396, "total_tokens": 450, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 35, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008712, "usd": 0.0012875657142857144}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "09ae7b9de7fd60f6d2f751e77180dac1d29b7e7a4b5bae0e07b55345035dd761"}, "error": null}
|
||||
{"id": "train-ja-006", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "固定ができない･･･なんか斜めになる ズームするために回したら望遠本体まで動くから固定できず写真が撮れない。あと三脚なら固定できるだろうと思ってやったらできなかった…。魚眼レンズとかは全然使えるけど1番使いたかった望遠レンズが使えなくて残念。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8782a5c09dd52de39324", "response_model": "kimi-k3", "response_created": 1785366403, "request_started_at": "2026-07-29T23:06:42.367221+00:00", "latency_seconds": 2.3845060421153903, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese, which is outside listed set => ot. Need final exactly one line. Ensure no extra.", "prediction": "ot", "usage": {"completion_tokens": 43, "prompt_tokens": 439, "total_tokens": 482, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 24, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008472, "usd": 0.0012520955844155847}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e94e46debcc43209a2807393b6cc60c7e7ed14a63c6aa34bdd2b7feff187c2ec"}, "error": null}
|
||||
{"id": "train-ja-007", "split": "train", "source_label": "ja", "gold_label": "ot", "text": "ずっと前から欲しかったですが、高価なイメージがあって。今回、充電の持ちと手頃なお値段で、こちらを選びました！ペアリングもスムーズにできたし、軽いし、耳へのフィット感も音もすごくいいです。ちなみに電子レンジ使う時だけ（特にレンジの使い始め）たまに途切れます。しかし、全然許容範囲です。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8783bc7cc35e30f1df13", "response_model": "kimi-k3", "response_created": 1785366403, "request_started_at": "2026-07-29T23:06:42.532607+00:00", "latency_seconds": 3.4559877500869334, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese. Japanese is not in the listed labels, so it should be classified as ot. The script is Japanese (hiragana/katakana/kanji mix), clearly not Chinese (uses kana). Since Japanese is outside the listed set, answer ot.", "prediction": "ot", "usage": {"completion_tokens": 76, "prompt_tokens": 470, "total_tokens": 546, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 57, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012392, "usd": 0.001831441038961039}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "abdd556a3890021596512dc16ea2750483d99c213dd92e4c81f8388cef93c35f"}, "error": null}
|
||||
{"id": "train-nl-000", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Een hond die op een blikje bijt.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878389b4c0360c6466bf", "response_model": "kimi-k3", "response_created": 1785366404, "request_started_at": "2026-07-29T23:06:42.821122+00:00", "latency_seconds": 3.407398040872067, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Dutch: \"Een hond die op een blikje bijt.\" Not in listed labels, so ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 53, "prompt_tokens": 357, "total_tokens": 410, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 34, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007832, "usd": 0.0011575085714285716}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5280586a519121d22b69cb7cde464c244a3b7ad9b1aaa83f07d70f3c452d0c5c"}, "error": null}
|
||||
{"id": "train-nl-001", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Rusland zegt dat ballistische raketten afgevuurd zijn in het Middellandse Zeegebied...", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878443d59830337427bb", "response_model": "kimi-k3", "response_created": 1785366405, "request_started_at": "2026-07-29T23:06:42.861440+00:00", "latency_seconds": 4.365280167199671, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Dutch: \"Rusland zegt dat ballistische raketten afgevuurd zijn in het Middellandse Zeegebied...\" Dutch is not in the listed labels, so classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 67, "prompt_tokens": 372, "total_tokens": 439, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 48, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009532, "usd": 0.0014087553246753249}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "cb45f47344d7125a18d40614f4734b58ccc29b19a3a5526637ff5b4185a6bd46"}, "error": null}
|
||||
{"id": "train-nl-002", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Iran atoombusbesprekingen beginnen in de hoop op vooruitgang...", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8783571e6ffa0b9caba9", "response_model": "kimi-k3", "response_created": 1785366404, "request_started_at": "2026-07-29T23:06:43.114847+00:00", "latency_seconds": 3.9983460842631757, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"Iran atoombusbesprekingen beginnen in de hoop op vooruitgang...\" contains Dutch words: \"atoombusbesprekingen\" (nuclear talks), \"beginnen\" (begin), \"in de hoop op vooruitgang\" (in the hope of progress). Dutch is not in the listed labels, so classify as ot. Need final exact line.", "prediction": "ot", "usage": {"completion_tokens": 102, "prompt_tokens": 363, "total_tokens": 465, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 83, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012852, "usd": 0.0018994254545454548}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5b29076d24041e0bfd84e91d53d4c4353d1f9c67f9860b73bd2744fa2407f0c8"}, "error": null}
|
||||
{"id": "train-nl-003", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Twee vrouwen die op de bruine bank zitten.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8784e3dfcfab6fd1019e", "response_model": "kimi-k3", "response_created": 1785366405, "request_started_at": "2026-07-29T23:06:44.351441+00:00", "latency_seconds": 4.74173391610384, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user wants language classification. Text: \"Twee vrouwen die op de bruine bank zitten.\" This is Dutch, not in list (de is German but this is Dutch). Words: Twee (two), vrouwen (women), die (who), op de (on the), bruine (brown), bank (couch), zitten (sit). Clearly Dutch. Not among listed labels, so ot. Need final exactly one line. Ensure no extra. Already used. Final concise.", "prediction": "ot", "usage": {"completion_tokens": 125, "prompt_tokens": 359, "total_tokens": 484, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 106, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015072, "usd": 0.002227524155844156}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3a030dfeadd80dd890bb123ad9ba9156baf4ab6668c4dfcc68c786a708f9b831"}, "error": null}
|
||||
{"id": "train-nl-004", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Er loopt een hond in de sneeuw.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87858b578eddec83028d", "response_model": "kimi-k3", "response_created": 1785366406, "request_started_at": "2026-07-29T23:06:44.631800+00:00", "latency_seconds": 3.6072416668757796, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"Er loopt een hond in de sneeuw.\" is Dutch. Dutch is not in the listed labels, so classify as ot. Final answer exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 57, "prompt_tokens": 356, "total_tokens": 413, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 38, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008212, "usd": 0.0012136696103896105}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "da764077791151ca686d846fe5dfbcc60f545cf689fc686531434c6d81717782"}, "error": null}
|
||||
{"id": "train-nl-005", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Freddie Starr gearresteerd in Savile Abuse Sonde...", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8785ec03bada3b15cb48", "response_model": "kimi-k3", "response_created": 1785366406, "request_started_at": "2026-07-29T23:06:44.734351+00:00", "latency_seconds": 4.4337359997443855, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"Freddie Starr gearresteerd in Savile Abuse Sonde...\" is a mix. \"Freddie Starr\" English proper noun, \"gearresteerd\" Dutch for arrested, \"in\" could be Dutch/English, \"Savile Abuse\" English, \"Sonde\" German/Dutch for probe/inquiry. This looks Dutch/English mixed headline. Not in listed languages (Dutch not listed) or deliberate mixed. Classify ot. Need final exactly.", "prediction": "ot", "usage": {"completion_tokens": 122, "prompt_tokens": 360, "total_tokens": 482, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 103, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014792, "usd": 0.002186142337662338}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "54fd64a8772b85d039c53a2d94c4ec3df256d62e9dfd534c5c7af279afe6693c"}, "error": null}
|
||||
{"id": "train-nl-006", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Een hond in een auto.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8785124dbf4a23c98923", "response_model": "kimi-k3", "response_created": 1785366405, "request_started_at": "2026-07-29T23:06:44.752481+00:00", "latency_seconds": 2.1942561669275165, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Dutch \"Een hond in een auto\" — outside listed set → ot.", "prediction": "ot", "usage": {"completion_tokens": 38, "prompt_tokens": 352, "total_tokens": 390, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 19, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006232, "usd": 0.000921041038961039}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e0915cf38dbec2bdadcad1c6f71827f4d3298a40e74b2fd74a19409d8325bf3c"}, "error": null}
|
||||
{"id": "train-nl-007", "split": "train", "source_label": "nl", "gold_label": "ot", "text": "Een aanrecht en aanrechtblad met schalen op de planken.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878578b4e9389290f817", "response_model": "kimi-k3", "response_created": 1785366406, "request_started_at": "2026-07-29T23:06:44.990072+00:00", "latency_seconds": 2.6873428751714528, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Dutch, not listed → ot.", "prediction": "ot", "usage": {"completion_tokens": 28, "prompt_tokens": 361, "total_tokens": 389, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 9, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005412, "usd": 0.0007998514285714286}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "396d629ef9b269976d41196bd2f9ff6f846e637b365d8e1b19db748ea01922de"}, "error": null}
|
||||
{"id": "train-pl-000", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Dwie dziewczyny z kucykami jeżdżą na przejażdżce w parku rozrywki.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8786a5f8e9319959d631", "response_model": "kimi-k3", "response_created": 1785366407, "request_started_at": "2026-07-29T23:06:45.611954+00:00", "latency_seconds": 4.201343417167664, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Polish: \"Dwie dziewczyny z kucykami jeżdżą na przejażdżce w parku rozrywki.\" Polish is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 77, "prompt_tokens": 376, "total_tokens": 453, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 58, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010612, "usd": 0.0015683709090909092}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b8e76f56558c349be6c6e3287e591b2d21883d97e42612f872e75cb7414ef54a"}, "error": null}
|
||||
{"id": "train-pl-001", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "alstom konkuruje o kontrakt z krajami japońskimi i niemieckimi.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8786d1bc7b6a3fb5f732", "response_model": "kimi-k3", "response_created": 1785366407, "request_started_at": "2026-07-29T23:06:45.989253+00:00", "latency_seconds": 2.5491006672382355, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Polish. Not in list → ot.", "prediction": "ot", "usage": {"completion_tokens": 29, "prompt_tokens": 368, "total_tokens": 397, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 10, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005652, "usd": 0.0008353215584415585}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0d2ac9e54665bce7dfac6f0a81333b46c6bbbdbefebbe4f97124e046a8512053"}, "error": null}
|
||||
{"id": "train-pl-002", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Pankda je bambus.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8786f4fe3c88d68d6911", "response_model": "kimi-k3", "response_created": 1785366407, "request_started_at": "2026-07-29T23:06:46.229492+00:00", "latency_seconds": 5.357525875326246, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Text: 'Pankda je bambus.' Latin script. Words: 'Pankda' unknown, 'je' could be Slovenian/Croatian/Serbian 'is', 'bambus' bamboo in Slovenian/Croatian/etc. Looks Slovenian-ish sentence 'Pankda je bambus' maybe 'Panda je bambus' misspelled? Language outside listed set likely Slovenian -> ot. Not Turkish/German etc. Very short but signal 'je bambus' Slavic/Balkan not listed. Label ot.", "prediction": "ot", "usage": {"completion_tokens": 136, "prompt_tokens": 352, "total_tokens": 488, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 117, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016032, "usd": 0.002369404675324676}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "04f170239043965a580280ca7d33bf5624e4504fafb3db289b4d16f813450494"}, "error": null}
|
||||
{"id": "train-pl-003", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Libijczycy zaczynają oddawać broń", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87865657fe29c59a3a6d", "response_model": "kimi-k3", "response_created": 1785366407, "request_started_at": "2026-07-29T23:06:46.418147+00:00", "latency_seconds": 3.247009541839361, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Polish: \"Libijczycy zaczynają oddawać broń\" means \"Libyans are starting to hand over weapons.\" Polish is outside the listed labels, so ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 71, "prompt_tokens": 361, "total_tokens": 432, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 52, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009712, "usd": 0.0014353579220779222}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "acfec58d9f435cac74ad108ebb0d66b2d1abe809b8d1ba5e6096e661b2040f53"}, "error": null}
|
||||
{"id": "train-pl-004", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "4 zagranicznych żołnierzy zabitych na wschodzie Afganistanu", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8787c6a3424af9907964", "response_model": "kimi-k3", "response_created": 1785366408, "request_started_at": "2026-07-29T23:06:46.421727+00:00", "latency_seconds": 5.408676082734019, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Polish: \"4 zagranicznych żołnierzy zabitych na wschodzie Afganistanu\" means \"4 foreign soldiers killed in eastern Afghanistan\". Polish is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 81, "prompt_tokens": 370, "total_tokens": 451, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 62, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010892, "usd": 0.0016097527272727276}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "62fb3400f67e6d75f6e344ef273d2b69de81966944403267f63df37e2cd98e58"}, "error": null}
|
||||
{"id": "train-pl-005", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Egipt głosuje nad nową konstytucją", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8787bc7cc35e30f1df18", "response_model": "kimi-k3", "response_created": 1785366408, "request_started_at": "2026-07-29T23:06:46.947477+00:00", "latency_seconds": 3.1497338749468327, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user is asking to classify text \"Egipt głosuje nad nową konstytucją\" - this is Polish, meaning \"Egypt votes on new constitution\". Polish is outside the listed set, so label is ot. Final Answer: ot.", "prediction": "ot", "usage": {"completion_tokens": 75, "prompt_tokens": 360, "total_tokens": 435, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 56, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010092, "usd": 0.0014915189610389611}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c81e1ddb8c863c08f0a3920b98c909e49a0bb19f3470e930c7fe4f832b597330"}, "error": null}
|
||||
{"id": "train-pl-006", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Dwie osoby grają w golfa na polu golfowym.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8787d1bc7b6a3fb5f735", "response_model": "kimi-k3", "response_created": 1785366408, "request_started_at": "2026-07-29T23:06:47.113864+00:00", "latency_seconds": 3.24384529190138, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Polish: \"Dwie osoby grają w golfa na polu golfowym.\" Polish is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 61, "prompt_tokens": 360, "total_tokens": 421, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 42, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008692, "usd": 0.0012846098701298702}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e4e23c8ae69d47802e9087618ae09c5949db0666eaf04bca5798322572325f2d"}, "error": null}
|
||||
{"id": "train-pl-007", "split": "train", "source_label": "pl", "gold_label": "ot", "text": "Nigdy się z tego nie wydostaliśmy!", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87886aa29639cda94016", "response_model": "kimi-k3", "response_created": 1785366409, "request_started_at": "2026-07-29T23:06:47.227220+00:00", "latency_seconds": 3.708410833030939, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is \"Nigdy się z tego nie wydostaliśmy!\" which is Polish for \"We never got out of it!\" Polish is not in the listed labels, so classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 64, "prompt_tokens": 359, "total_tokens": 423, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 45, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008972, "usd": 0.0013259916883116886}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "bd53e7f9e2e77f23513fda5744d4afef7b02165910caed484d3cf8c1d90f8a95"}, "error": null}
|
||||
{"id": "train-pt-000", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Combatentes rebeldes 'capturam' soldados sírios", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8788dd32e5734ad628f4", "response_model": "kimi-k3", "response_created": 1785366409, "request_started_at": "2026-07-29T23:06:47.677999+00:00", "latency_seconds": 7.808012790977955, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is \"Combatentes rebeldes 'capturam' soldados sírios\". Words: Combatentes (fighters), rebeldes (rebels), capturam (capture), soldados (soldiers), sírios (Syrian). This is Portuguese vocabulary/function words. Portuguese is not in listed labels (ar,de,el,en,es,fr,hi,ru,tr,ur,vi,zh). It's Latin script and could be confused with Spanish, but \"Combatentes\", \"rebeldes\", \"capturam\", \"sírios\" are Portuguese; Spanish would be \"combatientes rebeldes capturan soldados sirios\". Since pt is outside listed set and not deliberate mixed among listed, label ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 178, "prompt_tokens": 358, "total_tokens": 536, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 159, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.020352, "usd": 0.003007867012987013}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "633b4170364ace5b5a242b7d332f6f16bf076d58e0051dd6c7c072e54c2cbfe7"}, "error": null}
|
||||
{"id": "train-pt-001", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Nenhum acordo sobre penhasco fiscal enquanto Obama vai de férias", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878812e92b55023aa891", "response_model": "kimi-k3", "response_created": 1785366409, "request_started_at": "2026-07-29T23:06:48.239754+00:00", "latency_seconds": 3.617832582909614, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Portuguese: \"Nenhum acordo sobre penhasco fiscal enquanto Obama vai de férias\" means \"No deal on fiscal cliff while Obama goes on vacation.\" Portuguese is not in the listed labels, so classify as ot. Need final exact line only.", "prediction": "ot", "usage": {"completion_tokens": 78, "prompt_tokens": 364, "total_tokens": 442, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 59, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010472, "usd": 0.0015476800000000003}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "63f253e7d0802f4ee4c6ca4e0cabc45b1872149085c822b1bdf45ddfb5d34f2c"}, "error": null}
|
||||
{"id": "train-pt-002", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Não foi possível contactar imediatamente representantes seqüenciais para comentários sobre o anúncio da SCO.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878931fa7ecf69e5f7fd", "response_model": "kimi-k3", "response_created": 1785366410, "request_started_at": "2026-07-29T23:06:48.538936+00:00", "latency_seconds": 4.370871542021632, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user wants me to classify the language. The text is \"Não foi possível contactar imediatamente representantes seqüenciais para comentários sobre o anúncio da SCO.\" This is Portuguese. Portuguese is not in the listed labels, so it should be \"ot\". Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 89, "prompt_tokens": 375, "total_tokens": 464, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 70, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011792, "usd": 0.0017427657142857146}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "69c072c8ea7075ae0cf4c867dc5880544efaf0ea5192b451efaf94ea2ed30e35"}, "error": null}
|
||||
{"id": "train-pt-003", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Quatro aviões azuis e amarelos sobrevoando quatro barcos.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8789419b7910c2fd95f1", "response_model": "kimi-k3", "response_created": 1785366410, "request_started_at": "2026-07-29T23:06:49.094104+00:00", "latency_seconds": 3.204384167212993, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Text: \"Quatro aviões azuis e amarelos sobrevoando quatro barcos.\" This is Portuguese, not listed among labels, so ot. Need final exact.", "prediction": "ot", "usage": {"completion_tokens": 59, "prompt_tokens": 364, "total_tokens": 423, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 40, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008572, "usd": 0.0012668748051948052}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f279b9de88f91ab309478855bac0a7e5632b16ba8b7cd0a0dd26a4a0e63f6383"}, "error": null}
|
||||
{"id": "train-pt-004", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "O Comité Bancário do Senado está agendado para realizar uma audiência na terça-feira, onde Donaldson está agendado para testemunhar sobre hedge e fundos mútuos.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878b776710264cf67cde", "response_model": "kimi-k3", "response_created": 1785366411, "request_started_at": "2026-07-29T23:06:49.168766+00:00", "latency_seconds": 4.5320982919074595, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Portuguese: \"O Comité Bancário do Senado está agendado...\" Portuguese is not among listed labels, so classify as ot (other). Need final exactly line. Ensure no extra.", "prediction": "ot", "usage": {"completion_tokens": 63, "prompt_tokens": 391, "total_tokens": 454, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 44, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009512, "usd": 0.0014057994805194807}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "4ee2429b5746a7c4854b81efe51b165cb423fce7d3bdd2e7ccb90079a462b449"}, "error": null}
|
||||
{"id": "train-pt-005", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Thomas Cook acusado de colocar os custos à frente dos clientes", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878ab983a70a59cd3bf4", "response_model": "kimi-k3", "response_created": 1785366411, "request_started_at": "2026-07-29T23:06:49.666038+00:00", "latency_seconds": 4.030519540887326, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text: \"Thomas Cook acusado de colocar os custos à frente dos clientes\" is Portuguese. Portuguese is not in listed labels; rule says language outside listed set is ot. Need final exactly. Ensure no extra. Label ot.", "prediction": "ot", "usage": {"completion_tokens": 73, "prompt_tokens": 363, "total_tokens": 436, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 54, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009952, "usd": 0.0014708280519480522}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7c14c79f9eda71f8ec7707feecdae1e4522e339ba902ebc1bbfe5bea9743ebaf"}, "error": null}
|
||||
{"id": "train-pt-006", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Um cão corre à volta de um quintal.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878bc33554b59be5be88", "response_model": "kimi-k3", "response_created": 1785366412, "request_started_at": "2026-07-29T23:06:49.814041+00:00", "latency_seconds": 4.387702708598226, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"Um cão corre à volta de um quintal.\" is Portuguese. Portuguese is not in the listed labels, so it should be classified as \"ot\" (other languages).", "prediction": "ot", "usage": {"completion_tokens": 58, "prompt_tokens": 355, "total_tokens": 413, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 39, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008292, "usd": 0.0012254929870129873}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "028f4a4151fce7cc14b6d6086b6b9cd7702122a863205d014c7d0a39936dc093"}, "error": null}
|
||||
{"id": "train-pt-007", "split": "train", "source_label": "pt", "gold_label": "ot", "text": "Um homem está a saltar uma parede.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878b55543c615b08135a", "response_model": "kimi-k3", "response_created": 1785366412, "request_started_at": "2026-07-29T23:06:50.097886+00:00", "latency_seconds": 4.741404332686216, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user wants me to classify the language of \"Um homem está a saltar uma parede.\" This is Portuguese, which is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh). So it should be classified as ot (other).", "prediction": "ot", "usage": {"completion_tokens": 87, "prompt_tokens": 355, "total_tokens": 442, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 68, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011192, "usd": 0.00165409038961039}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0367a8e77078e45b8d9a0dd4de9cbe3b2ac79fceeff4d67055b861dbb9d59ba4"}, "error": null}
|
||||
{"id": "train-ru-000", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Nsiad использует выводы таким образом в рамках своей текущей работы по двусторонним инициативам .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878a80beffa1ad578f7a", "response_model": "kimi-k3", "response_created": 1785366411, "request_started_at": "2026-07-29T23:06:50.358444+00:00", "latency_seconds": 4.773756290785968, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is in Cyrillic: \"Nsiad использует выводы таким образом в рамках своей текущей работы по двусторонним инициативам.\" Words like использует, выводы, таким образом, в рамках, своей текущей работы, по двусторонним инициативам are clearly Russian. \"Nsiad\" is a proper noun/acronym in Latin but ignore proper nouns. Label ru. Need output exactly one line.", "prediction": "ru", "usage": {"completion_tokens": 133, "prompt_tokens": 379, "total_tokens": 512, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 114, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016272, "usd": 0.0024048748051948053}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "37a91a647a604248ae27c49cbd8e25bffe0b80e971d06d7a2c820e963b6588de"}, "error": null}
|
||||
{"id": "train-ru-001", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Но это также о том времени , когда ты начинаешь получать температуру в кабине .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878c3393bfb96f2e6886", "response_model": "kimi-k3", "response_created": 1785366413, "request_started_at": "2026-07-29T23:06:50.936388+00:00", "latency_seconds": 4.909779625013471, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The user wants a one-line answer only. The text is Russian Cyrillic: \"Но это также о том времени , когда ты начинаешь получать температуру в кабине .\" Clearly Russian. Need final exactly: Final Answer: ru.", "prediction": "ru", "usage": {"completion_tokens": 76, "prompt_tokens": 370, "total_tokens": 446, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 57, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010392, "usd": 0.0015358566233766235}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "32e613803024571606431a3ed6ff105072e1510e2d60e3caa9b674d521489026"}, "error": null}
|
||||
{"id": "train-ru-002", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Поощряется проведение экспериментов , равно как и представление такой дополнительной информации , что будет способствовать укреплению финансового доклада .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878c12e92b55023aa899", "response_model": "kimi-k3", "response_created": 1785366413, "request_started_at": "2026-07-29T23:06:51.588054+00:00", "latency_seconds": 4.48610133305192, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is in Cyrillic script with Russian vocabulary and grammar: \"Поощряется проведение экспериментов\" etc. Clearly Russian. Need output exactly Final Answer: ru. Ensure final only one line.", "prediction": "ru", "usage": {"completion_tokens": 67, "prompt_tokens": 394, "total_tokens": 461, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 48, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009972, "usd": 0.0014737838961038962}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "083f2365746397d1afa1e4f873f46b41dc26f41f856c7ce86fcc4766b7b75080"}, "error": null}
|
||||
{"id": "train-ru-003", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Да , это совсем другая культура . Это странно внизу , потому что", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878da770d98a4bce25ee", "response_model": "kimi-k3", "response_created": 1785366413, "request_started_at": "2026-07-29T23:06:51.831217+00:00", "latency_seconds": 3.987691250164062, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is Cyrillic: \"Да, это совсем другая культура. Это странно внизу, потому что\" — clearly Russian vocabulary (да, это, совсем, другая, культура, странно, потому что). Label ru.", "prediction": "ru", "usage": {"completion_tokens": 81, "prompt_tokens": 369, "total_tokens": 450, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 62, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010872, "usd": 0.0016067968831168832}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "acdeea1a3f1a07ada687608e47bbdf01da099aacff1a115ce14cbcabf4353163"}, "error": null}
|
||||
{"id": "train-ru-004", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Присяжные посмотрели вверх , заинтересовались .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878c697b682b78094377", "response_model": "kimi-k3", "response_created": 1785366413, "request_started_at": "2026-07-29T23:06:51.858494+00:00", "latency_seconds": 4.619346875231713, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is in Russian Cyrillic: \"Присяжные посмотрели вверх, заинтересовались.\" Means \"The jurors looked up, became interested.\" Label ru. Final only.", "prediction": "ru", "usage": {"completion_tokens": 65, "prompt_tokens": 363, "total_tokens": 428, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 46, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009152, "usd": 0.001352594285714286}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "62ea29e3236b96149da78e7a593ab2782aeeaa237701ca23d0de99cec230ca2e"}, "error": null}
|
||||
{"id": "train-ru-005", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Но может ли он навсегда ?", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878c5657fe29c59a3a7a", "response_model": "kimi-k3", "response_created": 1785366413, "request_started_at": "2026-07-29T23:06:52.299471+00:00", "latency_seconds": 4.193642290774733, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is Russian Cyrillic: \"Но может ли он навсегда ?\" translates to \"But can he forever?\" Clearly Russian. Need output exactly Final Answer: ru.", "prediction": "ru", "usage": {"completion_tokens": 58, "prompt_tokens": 355, "total_tokens": 413, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 39, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008292, "usd": 0.0012254929870129873}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "d64ef80a24739cf035739c5da72727a027c73ab7ad8ef8d97d73795dbbc4352f"}, "error": null}
|
||||
{"id": "train-ru-006", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "И я могу назвать несколько больше , я просто не могу придумать их имена прямо с рук .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878ec85bf7c1921a6e62", "response_model": "kimi-k3", "response_created": 1785366415, "request_started_at": "2026-07-29T23:06:52.910636+00:00", "latency_seconds": 4.6627399157732725, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text uses Cyrillic script and Russian words: \"И я могу назвать несколько больше, я просто не могу придумать их имена прямо с рук.\" This is Russian. Final label ru.", "prediction": "ru", "usage": {"completion_tokens": 72, "prompt_tokens": 377, "total_tokens": 449, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 53, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010132, "usd": 0.0014974306493506495}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8fc69d380af0af3945550292a493019ffede0a47efc62c64199db916d629a049"}, "error": null}
|
||||
{"id": "train-ru-007", "split": "train", "source_label": "ru", "gold_label": "ru", "text": "Межучрежденческой совет по делам бездомных находится на ранних этапах своего 10-летнего плана по ликвидации бездомности , как это было сказано .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878ea5c09dd52de39331", "response_model": "kimi-k3", "response_created": 1785366414, "request_started_at": "2026-07-29T23:06:53.697226+00:00", "latency_seconds": 3.211011374834925, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is in Cyrillic script with Russian vocabulary: \"Межучрежденческой совет по делам бездомных находится на ранних этапах своего 10-летнего плана по ликвидации бездомности\". Clearly Russian. Label ru.", "prediction": "ru", "usage": {"completion_tokens": 84, "prompt_tokens": 397, "total_tokens": 481, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 65, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011732, "usd": 0.0017338981818181818}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "fc3686d4ae70b4e378732640daf5a1feabaf3b9b24893e74c373addae6ab14b8"}, "error": null}
|
||||
{"id": "train-sw-000", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Isipokuwa kubwa ni uandishi wa habari wa ufaransa , ambapo heshima wakati kwa mitindo imekuwa ni mtazamo wa kawaida tangu karne ya 17", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878ea770d98a4bce25ef", "response_model": "kimi-k3", "response_created": 1785366414, "request_started_at": "2026-07-29T23:06:53.701423+00:00", "latency_seconds": 3.4830967909656465, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text appears to be Swahili. \"Isipokuwa\", \"kubwa\", \"uandishi wa habari\", \"ambapo\", \"imekuwa\", \"tangu karne\" are Swahili words. Swahili is not in the listed labels, so it should be classified as ot.", "prediction": "ot", "usage": {"completion_tokens": 87, "prompt_tokens": 390, "total_tokens": 477, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 68, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011892, "usd": 0.0017575449350649352}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8ece726f2b1e652864c9a126e368af52455e3dd8127064256473ab535a81d66a"}, "error": null}
|
||||
{"id": "train-sw-001", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Mipangilio ya uga .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878eaff9f7f7f84cf451", "response_model": "kimi-k3", "response_created": 1785366415, "request_started_at": "2026-07-29T23:06:54.202344+00:00", "latency_seconds": 3.663800540845841, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user wants a final exact line. Need classify text \"Mipangilio ya uga .\" Looks Swahili: Mipangilio = arrangements/settings, ya = of, uga maybe courtyard? Swahili not in listed labels => ot. Need output exactly Final Answer: ot. Ensure no extra. Done.", "prediction": "ot", "usage": {"completion_tokens": 86, "prompt_tokens": 351, "total_tokens": 437, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 67, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011012, "usd": 0.0016274877922077924}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "4b15fbdb4605e6fb0debc6a9cf462f8fb3cc1ac99256f9f72a77b98d29c2798d"}, "error": null}
|
||||
{"id": "train-sw-002", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Ofisi ya usimamizi wa habari ( oim ) , ofisi ya utendaji wa programu ( opp ) , na timu ya kupanga ya serikali sasa ni wafanyakazi .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a878fe3dfcfab6fd101a9", "response_model": "kimi-k3", "response_created": 1785366415, "request_started_at": "2026-07-29T23:06:54.839648+00:00", "latency_seconds": 3.2575742499902844, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Swahili: \"Ofisi ya usimamizi wa habari... ofisi ya utendaji wa programu... timu ya kupanga ya serikali sasa ni wafanyakazi.\" Swahili is not in listed labels, so ot. Need final exact one line.", "prediction": "ot", "usage": {"completion_tokens": 85, "prompt_tokens": 391, "total_tokens": 476, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 66, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011712, "usd": 0.0017309423376623378}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "60edf864a2a8989b0833e6a2070e0c359a26c035c5fb8f010314134d3fd04124"}, "error": null}
|
||||
{"id": "train-sw-003", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Nguvu hii ya asili ana na mwanaume tangu mara ya mapema , kama kuonyesha na pango-hekalu la pan , kwa ambaye raia wa syria na greeks kujitolea mkondo huo .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879032ee9d9a487694f0", "response_model": "kimi-k3", "response_created": 1785366417, "request_started_at": "2026-07-29T23:06:55.132846+00:00", "latency_seconds": 4.926183166913688, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Latin script, clearly Swahili-like: \"Nguvu hii ya asili ana na mwanaume tangu mara ya mapema, kama kuonyesha...\" Words like hii, ya, na, tangu, kama, kwa, ambaye are Bantu/Swahili. Swahili is not in listed labels, so classify ot. Need final exactly line.", "prediction": "ot", "usage": {"completion_tokens": 106, "prompt_tokens": 403, "total_tokens": 509, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 87, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014052, "usd": 0.002076776103896104}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "cd62ca609a547e43ff6fd27d0965958f6cf3f7a470a401645ac2dfa0ef5cf5a8"}, "error": null}
|
||||
{"id": "train-sw-004", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Mawasiliano ya kibinafsi ( 2 ) pamoja na ande salimbot , undugu wa kimataifa wa boilermakers , wajenzi wa meli , blacksmiths , wazushi na wasaidizi , februari 22 , 2002 .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879001c438c1be93d279", "response_model": "kimi-k3", "response_created": 1785366417, "request_started_at": "2026-07-29T23:06:55.486801+00:00", "latency_seconds": 4.679569417145103, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Swahili. Words like \"Mawasiliano ya kibinafsi\", \"pamoja na\", \"undugu wa kimataifa\", \"wajenzi wa meli\", \"wazushi na wasaidizi\" are clearly Swahili. Swahili is not in the listed labels, so classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 94, "prompt_tokens": 408, "total_tokens": 502, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 75, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012952, "usd": 0.0019142046753246756}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0357d8f4fa5784d15d57d755b55d9c5457181cf03ca74186fa04c406f8bce67b"}, "error": null}
|
||||
{"id": "train-sw-005", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Watu nani haki zao na kila kitu ng ' ambo na kila kitu na i think uh sijui kwa hiyo", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8790e5cd1f53d10ec53e", "response_model": "kimi-k3", "response_created": 1785366416, "request_started_at": "2026-07-29T23:06:55.819736+00:00", "latency_seconds": 2.9998759999871254, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text mixes Swahili words (Watu, haki zao, kila kitu, sijui, kwa hiyo) with English (\"i think\"). Rule 4: deliberate mixed-language is ot. Need final exact format.", "prediction": "ot", "usage": {"completion_tokens": 70, "prompt_tokens": 377, "total_tokens": 447, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 51, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009932, "usd": 0.001467872207792208}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "67acb2836dbfce2c1efc3927f9a11edd2de4d808377d4115eabe58a9dfa7e58d"}, "error": null}
|
||||
{"id": "train-sw-006", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Kuunda mazingira ya ushindani wange , kwa kiwango cha chini , kuondoa statutes4 ya kibinafsi na kanuni ya barua pepe .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8790f533e66c8281cad0", "response_model": "kimi-k3", "response_created": 1785366417, "request_started_at": "2026-07-29T23:06:55.846916+00:00", "latency_seconds": 3.2657720833085477, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text appears to be Swahili: \"Kuunda mazingira ya ushindani...\" Swahili is not in the listed labels, so it should be classified as \"ot\" (other).", "prediction": "ot", "usage": {"completion_tokens": 63, "prompt_tokens": 385, "total_tokens": 448, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 44, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009392, "usd": 0.0013880644155844157}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "42547adb51c0291567fdc037abe2330d65afbdfdc8040dc062bd1ebc47f0ad8c"}, "error": null}
|
||||
{"id": "train-sw-007", "split": "train", "source_label": "sw", "gold_label": "ot", "text": "Akiba ya kitaifa ya kitaifa", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87914c7866f47529520a", "response_model": "kimi-k3", "response_created": 1785366418, "request_started_at": "2026-07-29T23:06:56.074421+00:00", "latency_seconds": 4.008849000092596, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"Akiba ya kitaifa ya kitaifa\" is Swahili. \"Akiba\" means reserve/savings, \"ya kitaifa\" means national. Swahili is not in the listed labels, so classify as ot (other). Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 80, "prompt_tokens": 355, "total_tokens": 435, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 61, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010492, "usd": 0.0015506358441558443}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a878484d90176f9b0b38c3979f99c1479d05549e79bc61960c6d307e50fa9501"}, "error": null}
|
||||
{"id": "train-th-000", "split": "train", "source_label": "th", "gold_label": "ot", "text": "โอ้ มัน ฟัง ดู ยอดเยี่ยม", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87900b8d682056dbefad", "response_model": "kimi-k3", "response_created": 1785366417, "request_started_at": "2026-07-29T23:06:56.478563+00:00", "latency_seconds": 2.9028203329071403, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Thai script: \"โอ้ มัน ฟัง ดู ยอดเยี่ยม\" which means \"Oh, that sounds great\" in Thai. Thai is not in the listed labels, so it should be classified as \"ot\" (all other languages / unknown).", "prediction": "ot", "usage": {"completion_tokens": 86, "prompt_tokens": 366, "total_tokens": 452, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 67, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011312, "usd": 0.0016718254545454545}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "bb4bbb03387ea577da2e73b260770c987e356c3ab51b8be79f10cf83ac4b656a"}, "error": null}
|
||||
{"id": "train-th-001", "split": "train", "source_label": "th", "gold_label": "ot", "text": "ขอบใจ หลาย ๆ เด้อ รี เขา จากไป ด้วย การ เดิน แบบ เดียว กับ ที่ เขา มี จาก ซอย ที่ มั่นคง", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8791419b7910c2fd95f8", "response_model": "kimi-k3", "response_created": 1785366417, "request_started_at": "2026-07-29T23:06:56.493671+00:00", "latency_seconds": 4.10671945894137, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Thai script: \"ขอบใจ หลาย ๆ เด้อ...\" Thai is not among the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh), so it should be classified as ot (all other languages). Need output exactly final line. Ensure no extra.", "prediction": "ot", "usage": {"completion_tokens": 97, "prompt_tokens": 412, "total_tokens": 509, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 78, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013332, "usd": 0.0019703657142857143}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b6bbdb35091d2e426e17a28d3cfc1ab21c72768a5d2d774450d5f0369b988b21"}, "error": null}
|
||||
{"id": "train-th-002", "split": "train", "source_label": "th", "gold_label": "ot", "text": "แต่ ฉัน เดา ว่า", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8791be759e5d2a00e630", "response_model": "kimi-k3", "response_created": 1785366418, "request_started_at": "2026-07-29T23:06:56.908989+00:00", "latency_seconds": 3.186476457864046, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Thai script: \"แต่ ฉัน เดา ว่า\" which means \"but I guess that\". Thai is not in the listed labels, so it should be classified as ot (other languages).", "prediction": "ot", "usage": {"completion_tokens": 69, "prompt_tokens": 358, "total_tokens": 427, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 50, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009452, "usd": 0.0013969319480519483}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "99cb67fee88bbd029012fef05d0bee3157f5fa4fcd4656046d34b9cd73623399"}, "error": null}
|
||||
{"id": "train-th-003", "split": "train", "source_label": "th", "gold_label": "ot", "text": "ใช่ พวกเขา ควร ทำความสะอาด ให้ ชัดเจน ว่า มัน จะ ไม่ พา พวกเขา มาก ไป ใส่ สแตมป์ บน กระป๋อง น้ำผลไม้ ง่ายๆ เหมือน โซดา", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8791e3dfcfab6fd101ab", "response_model": "kimi-k3", "response_created": 1785366418, "request_started_at": "2026-07-29T23:06:57.185206+00:00", "latency_seconds": 3.127037708181888, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Thai script. Thai is not among the listed languages (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh). According to rule 4, a language outside the listed set is ot. So classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 79, "prompt_tokens": 446, "total_tokens": 525, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 60, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012212, "usd": 0.001804838441558442}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "337272146db400740bd354050a1fba2fc15fdd9aeaa8abfdca64eba902d48f53"}, "error": null}
|
||||
{"id": "train-th-004", "split": "train", "source_label": "th", "gold_label": "ot", "text": "การ ตั้งครรภ์ ธรรมชาติ ได้รับ การ ยกระดับ โดย ระบบ ของ ยุโรป ที่ ล้ำสมัย ที่สุด ของ ชลบุรี ยังคง ดำเนินการ อยู่ ใน คลอง กลาง ที่ คุณ จะ เห็น บน ทาง ของ คุณ ใต้ ไป ยัง วี", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87924cb5f60e0c8446be", "response_model": "kimi-k3", "response_created": 1785366418, "request_started_at": "2026-07-29T23:06:57.573993+00:00", "latency_seconds": 2.509723291732371, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Thai script, which is not among the listed labels. Per rules, language outside listed set is ot. Need output exactly Final Answer: ot.", "prediction": "ot", "usage": {"completion_tokens": 53, "prompt_tokens": 479, "total_tokens": 532, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 34, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010272, "usd": 0.0015181215584415585}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "bcc8e8a77a1271b79b3694440b00529824d98d91d8b4ae06ce69a0901fd2fbec"}, "error": null}
|
||||
{"id": "train-th-005", "split": "train", "source_label": "th", "gold_label": "ot", "text": "Atrisk งานก่อสร้าง ที่ แท้จริง มี การแสดง โดย ผู้รับเหมา การค้า ภายใต้ สัญญา กับ ซ. ผู้ แล้ว กลายเป็น ผู้รับผิดชอบ เจ้าของ สำหรับ การ ก่อสร้าง หมายถึง และ วิธีการ และ การจัดส่ง ของ สิ่งอำนวยความสะดวก ที่ สมบูรณ์ ภายใน ขอบเขต ของ เจ้าของงาน สำหรับ ต้นทุน เวลา และ คุณภาพ", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87922d573972cd4cdff1", "response_model": "kimi-k3", "response_created": 1785366419, "request_started_at": "2026-07-29T23:06:57.866773+00:00", "latency_seconds": 4.220352707896382, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is primarily Thai script. Thai is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh). There is a leading \"Atrisk\" Latin token, but the bulk is Thai. Since Thai is outside the listed set, classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 95, "prompt_tokens": 549, "total_tokens": 644, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 76, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015872, "usd": 0.0023457579220779222}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7067aba85fdddbaf79baaae60d6779e525bdd8c37a91894c000b9acbc8b29988"}, "error": null}
|
||||
{"id": "train-th-006", "split": "train", "source_label": "th", "gold_label": "ot", "text": "มัน ไม่ รบกวน ฉัน เลย ฉัน ไม่ รู้สึก ว่า มัน เป็นการ ละเมิด ความเป็นส่วนตัว หรือ อะไร เลย", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879312e760f13afdab25", "response_model": "kimi-k3", "response_created": 1785366420, "request_started_at": "2026-07-29T23:06:58.097885+00:00", "latency_seconds": 4.574286832939833, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Thai script. Thai is not among the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh). Therefore it falls under ot (all other languages / unknown). Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 76, "prompt_tokens": 412, "total_tokens": 488, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 57, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011232, "usd": 0.0016600020779220781}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b7644e3eca5acb404ae182ce88a45097a5c513dedd4ac140d68071d7dfd999d9"}, "error": null}
|
||||
{"id": "train-th-007", "split": "train", "source_label": "th", "gold_label": "ot", "text": "นี่ คือ ผล งานชิ้นเอก ของ อารยธรรม ฝรั่งเศส ที่ มี รอย เปลี่ยน ใน ศตวรรษ ที่ 12 จาก สไตล์ โร มาน ส ์ แบบ ไม่มีสติ ของ โบสถ์ เริ่ม สู่ ความ แข็งแกร่ง มากขึ้น", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8793adf8be004bbe4b3d", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:06:58.819910+00:00", "latency_seconds": 4.045349958818406, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Thai script, which is not in the listed labels. Therefore classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 40, "prompt_tokens": 474, "total_tokens": 514, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 21, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008872, "usd": 0.0013112124675324676}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c79ce6199a8ec82d9bcc5533a518e60706ef2a192be9a1dd501d139c14f48d37"}, "error": null}
|
||||
{"id": "train-tr-000", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Ayda 20 zloti veya her ay 20 zloti taksitle uygun bir alışveriş seçeneği sunuyoruz .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8793a770d98a4bce25f4", "response_model": "kimi-k3", "response_created": 1785366420, "request_started_at": "2026-07-29T23:06:59.113327+00:00", "latency_seconds": 3.942530042026192, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "Turkish: \"Ayda 20 zloti veya her ay 20 zloti taksitle uygun bir alışveriş seçeneği sunuyoruz.\" Turkish words: ayda, veya, her ay, taksitle, uygun, bir, alışveriş, seçeneği, sunuyoruz. Diacritics ş, ç, ı. Label tr.", "prediction": "tr", "usage": {"completion_tokens": 108, "prompt_tokens": 378, "total_tokens": 486, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 89, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013752, "usd": 0.002032438441558442}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "fc0f1dcb9412f8d9de8c8db88c9b919d5c0b56ec3be97b1addcc4c0d81865915"}, "error": null}
|
||||
{"id": "train-tr-001", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Amerika ' nın zamanında geleceğini içgüdüsel olarak hissetti .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8794fabdd4e38cb710e0", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:06:59.382418+00:00", "latency_seconds": 4.131074042059481, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "Turkish text with diacritics and words.", "prediction": "tr", "usage": {"completion_tokens": 32, "prompt_tokens": 365, "total_tokens": 397, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 13, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005892, "usd": 0.0008707916883116883}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8c19845adbb1351c545471e57569dde72af7331fa16e0e45271cea6d7c780638"}, "error": null}
|
||||
{"id": "train-tr-002", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Santorini ve mykonos özellikle avrupa ve ABD ' den bir sürü tasarımcı giyim ve ayakkabı var .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87944d654f2ee1729b50", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:07:00.059300+00:00", "latency_seconds": 2.542211249936372, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "Turkish: \"ve, özellikle, bir sürü, var\", diacritics.", "prediction": "tr", "usage": {"completion_tokens": 44, "prompt_tokens": 382, "total_tokens": 426, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 25, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007432, "usd": 0.0010983916883116885}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5cda423d739d6260aba91fbbc72beb5c4efe9057dd17c850e2bfc70e717d356b"}, "error": null}
|
||||
{"id": "train-tr-003", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Ca ' daan , kara sürücüler karşılığında bir şey söyledi mi diye olabilir .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87944c7866f47529520f", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:07:00.083905+00:00", "latency_seconds": 4.123532915953547, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The text uses Turkish-specific characters like ı, ş, and vocabulary such as \"karşılığında\", \"bir şey söyledi mi diye\", \"sürücüler\". Though punctuation is odd and first token \"Ca ' daan\" weird, overall language signal is Turkish. Need answer exact format.", "prediction": "tr", "usage": {"completion_tokens": 89, "prompt_tokens": 375, "total_tokens": 464, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 70, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011792, "usd": 0.0017427657142857146}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0165495c000d3a9f082971e13f195abdbd1bfdc529d6f37f270c0640947713e1"}, "error": null}
|
||||
{"id": "train-tr-004", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Hareketleri dikkatlice ölçülür .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8794a5c09dd52de39336", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:07:00.084554+00:00", "latency_seconds": 3.855686084367335, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The user wants classification. Need only final line exact. Text: \"Hareketleri dikkatlice ölçülür .\" Turkish words: hareketleri (movements), dikkatlice (carefully), ölçülür (is measured). Turkish diacritics ö. Label tr. Ensure visible response exactly Final Answer: tr.", "prediction": "tr", "usage": {"completion_tokens": 96, "prompt_tokens": 358, "total_tokens": 454, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 77, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012152, "usd": 0.0017959709090909091}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "6f569edb557ebb573b73f03cbc54538ef33728c2fdd6154cd993f4c15d253877"}, "error": null}
|
||||
{"id": "train-tr-005", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Işte bunu söylediğin için mutluyum ve gerek yoktu ama o da aynı şekilde hissediyorum ben oturup bazı programları izliyorum ve uh ve bile dahil olan insanlar için utanıyorum bunu kendine neden yapıyorsun biliyorsun", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8795079009943e6a64fe", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:07:00.095642+00:00", "latency_seconds": 6.4184724586084485, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The user wants me to classify the language of the provided text. Looking at the text, it contains Turkish-specific characters like \"Işte\", \"söylediğin\", \"mutluyum\", \"gerek\", \"hissediyorum\", \"izliyorum\", \"utanıyorum\". The dotted/dotless I (Işte with capital I and ş), and words like \"için\", \"ama\", \"ben\", \"neden\", \"biliyorsun\" are clearly Turkish. This is unambiguously Turkish. Final answer: tr.", "prediction": "tr", "usage": {"completion_tokens": 140, "prompt_tokens": 423, "total_tokens": 563, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 121, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.017852, "usd": 0.002638386493506494}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "827fbfde54bef450675e74e664408dd42353fccd9913ee61f344ab882fc01800"}, "error": null}
|
||||
{"id": "train-tr-006", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Kazanan giriş , rock kendini", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87943f3c9b5f06007f13", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:07:00.167025+00:00", "latency_seconds": 4.069066249765456, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The text: \"Kazanan giriş , rock kendini\". Words: \"Kazanan\" Turkish means winner, \"giriş\" Turkish entry/login with ş, \"kendini\" Turkish reflexive. \"rock\" English loan. Contains Turkish-specific ş and vocabulary. Classify tr.", "prediction": "tr", "usage": {"completion_tokens": 84, "prompt_tokens": 356, "total_tokens": 440, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 65, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010912, "usd": 0.0016127085714285716}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "def0e2c79f10bf729d77f5490c218c9cc3879d1d92785324ccdf25c0617656b4"}, "error": null}
|
||||
{"id": "train-tr-007", "split": "train", "source_label": "tr", "gold_label": "tr", "text": "Örneğin , bir bilgisayar sisteminin güvenli olduğu ve tanımlanmış bir ortamda çalışmasına izin veren yazılı bir yetkilendirme .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879412e92b55023aa8b2", "response_model": "kimi-k3", "response_created": 1785366421, "request_started_at": "2026-07-29T23:07:00.313193+00:00", "latency_seconds": 4.756401875056326, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The text contains Turkish-specific characters like Ö, ğ, ı, ç and words: Örneğin, bir, bilgisayar, sisteminin, güvenli, olduğu, tanımlanmış, ortamda, çalışmasına, izin, veren, yazılı, yetkilendirme. Clearly Turkish. Need output exactly final line.", "prediction": "tr", "usage": {"completion_tokens": 100, "prompt_tokens": 388, "total_tokens": 488, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 81, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013152, "usd": 0.0019437631168831172}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "85c940e5a3183fe6c9092e739f5e8db262e5be8d22ddc436b96688fd9cf09092"}, "error": null}
|
||||
{"id": "train-ur-000", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "ماس ماس ( ماس ) نیٹ ورک کی بنیاد پر یہ اثر انداز کیا گیا ہے .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87958d20c5ab26768723", "response_model": "kimi-k3", "response_created": 1785366422, "request_started_at": "2026-07-29T23:07:00.601206+00:00", "latency_seconds": 4.46722354227677, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text is in Arabic script. Need distinguish Arabic vs Urdu. Contains Urdu-specific word کی and نیٹ ورک (network), بنیاد پر, کیا گیا ہے — Urdu function words and sentence structure. Letter ے? not present but کی and گیا ہے strongly Urdu. Label ur.", "prediction": "ur", "usage": {"completion_tokens": 83, "prompt_tokens": 370, "total_tokens": 453, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 64, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011092, "usd": 0.001639311168831169}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "78251446cf3a06d29004ed0fb693e4bf37fbc5ab52fa50ba40a5c0819ea32f95"}, "error": null}
|
||||
{"id": "train-ur-001", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "اس میں تاخیر کی جاتی ہے . \"", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8797708f7e2c9beb0097", "response_model": "kimi-k3", "response_created": 1785366424, "request_started_at": "2026-07-29T23:07:02.087793+00:00", "latency_seconds": 4.526082457974553, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text uses Arabic script with Urdu-specific construction: \"اس میں تاخیر کی جاتی ہے\" (delay is done in this). Contains Urdu words میں, کی, جاتی, ہے. Urdu-specific grammar. Label ur. Ensure final exact one line.", "prediction": "ur", "usage": {"completion_tokens": 76, "prompt_tokens": 356, "total_tokens": 432, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 57, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010112, "usd": 0.0014944748051948053}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e610289ecb4f13308e213865e3be8883a9a10c66873abedc08dd5d14e6b08b45"}, "error": null}
|
||||
{"id": "train-ur-002", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "اور ہم نے اس وقت تک ایک مدت معین کیا ہے جس کی وجہ سے اس سے ملاقات کی جاتی ہے تو ہم نے رات کے اوقات میں رات یا اس سے بھی کچھ کم نہیں کیا ۔ کوفتے کی تراکیب", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8797de34a86cafee51e1", "response_model": "kimi-k3", "response_created": 1785366425, "request_started_at": "2026-07-29T23:07:02.602176+00:00", "latency_seconds": 6.281153957825154, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text is in Arabic script with Urdu-specific letters: ہ (Urdu heh), ے (Urdu yeh), and words like اور, ہم نے, اس, تک, ایک, مدت, معین, کیا, وجہ, سے, ملاقات, رات, اوقات, نہیں, کچھ, کم, کوفتے, تراکیب. The use of ہ, ے, and vocabulary is distinctly Urdu. Classify ur.", "prediction": "ur", "usage": {"completion_tokens": 121, "prompt_tokens": 405, "total_tokens": 526, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 102, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015592, "usd": 0.0023043761038961043}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e2c6a37e0f23ec0aebfef8bc3c14599aec01af0a118da2e7c4a9b8973286d027"}, "error": null}
|
||||
{"id": "train-ur-003", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "پھر ہم نے وزن کیا اور ہم نے اس میں سے ہر ایک کو عاجز کر دیا اور ہم ان سے بیزار ہیں", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87975657fe29c59a3a90", "response_model": "kimi-k3", "response_created": 1785366423, "request_started_at": "2026-07-29T23:07:02.672842+00:00", "latency_seconds": 5.9087252910248935, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "Text is Urdu: script Arabic with Urdu-specific ہ، ے? Words: پھر ہم نے وزن کیا... Urdu function words ہم نے، میں سے، ہر ایک کو، ہیں. Distinguish Arabic vs Urdu: has Urdu-specific letters ہ (gol he), ے? phrase uses ہم نے/ہیں typical Urdu, no Arabic case endings. So ur.", "prediction": "ur", "usage": {"completion_tokens": 107, "prompt_tokens": 377, "total_tokens": 484, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 88, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013632, "usd": 0.0020147033766233766}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "1a9d31d506a823d951de54737732d14e8590f0128c507cd8657ee921247de546"}, "error": null}
|
||||
{"id": "train-ur-004", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "کچھ بھی نہیں ۔", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8797448d84fdba942cd4", "response_model": "kimi-k3", "response_created": 1785366424, "request_started_at": "2026-07-29T23:07:02.866006+00:00", "latency_seconds": 7.001432499848306, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text is \"کچھ بھی نہیں ۔\" which is Urdu script. It contains Urdu-specific phrasing: کچھ بھی نہیں means \"nothing\" / \"not anything\". The letters include ے? Actually text: کچھ بھی نہیں. Uses چھ aspirate, بھی, نہیں. This is clearly Urdu phrase. Arabic would not use چ, پ, گ, ے typically. Distinguish Urdu-specific letters: چ is used in Urdu/Persian not Arabic; بھی and نہیں are Urdu/Hindi words. Label ur. Need final exactly.", "prediction": "ur", "usage": {"completion_tokens": 139, "prompt_tokens": 350, "total_tokens": 489, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 120, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016292, "usd": 0.0024078306493506498}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "9433cff40ea9aa3a7576e108df0c6a41591d157b753cf9c726d25744831918e6"}, "error": null}
|
||||
{"id": "train-ur-005", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "جب میں نے آخری بار ایک سال میں پہلی بار ایسوسی ایشن سے اظہار کیا ، جس کی وجہ سے آپ نے ایک دوسرے کے پیمانے پر کام کرنے کی کوشش کی ، اس کے بعد ، دیگر ٹرانسپورٹ فرمیں کے ساتھ کام کرنے کے لئے .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8798c683626717c42351", "response_model": "kimi-k3", "response_created": 1785366425, "request_started_at": "2026-07-29T23:07:03.056491+00:00", "latency_seconds": 7.266671542078257, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text is in Arabic script. Need distinguish Arabic vs Urdu. Contains Urdu-specific words: میں, نے, آخری, بار, سال, پہلی, ایسوسی ایشن, اظہار, آپ, دوسرے, پیمانے, کام, کرنے, کوشش, کے, بعد, ٹرانسپورٹ, فرمیں, ساتھ, لئے. Urdu-specific letters like ے, ں? It includes ٹ (retroflex) in ٹرانسپورٹ, ے in نے/دوسرے/پیمانے/کے/لئے, and vocabulary is Urdu. Not Arabic. Label ur. Final exact.", "prediction": "ur", "usage": {"completion_tokens": 168, "prompt_tokens": 421, "total_tokens": 589, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 149, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.020612, "usd": 0.003046292987012987}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "702d7926752249cde47e5858ef4af27fb430df4e4cf0fff38f21530d71c77137"}, "error": null}
|
||||
{"id": "train-ur-006", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "یا آپ کا خیال ہے کہ آپ کو اس طرح کے کم از کم ایک بار نظر آنا چاہتے ہیں ؟", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87970b8d682056dbefb6", "response_model": "kimi-k3", "response_created": 1785366424, "request_started_at": "2026-07-29T23:07:03.514280+00:00", "latency_seconds": 5.082493041176349, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The user wants language classification. Need output exactly Final Answer: xx. Need reason privately in tool. Text Arabic script: \"یا آپ کا خیال ہے کہ آپ کو اس طرح کے کم از کم ایک بار نظر آنا چاہتے ہیں ؟\" Contains Urdu-specific letters/words: آپ، کا، خیال، ہے، کہ، کو، اس طرح، کم از کم (Persian phrase used in Urdu), چاہتے ہیں. Urdu-specific چاہتے, ہیں, کا, کو. Not Arabic. Label ur. Need final exactly one line.", "prediction": "ur", "usage": {"completion_tokens": 141, "prompt_tokens": 373, "total_tokens": 514, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 122, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016952, "usd": 0.0025053735064935064}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "13ef92d7e9b512db676bfe8ed5f59a1580f044af874c5612f70ae5bbf3c776e8"}, "error": null}
|
||||
{"id": "train-ur-007", "split": "train", "source_label": "ur", "gold_label": "ur", "text": "واقعی رئیل اسپورٹس", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8798953d559c44153a94", "response_model": "kimi-k3", "response_created": 1785366425, "request_started_at": "2026-07-29T23:07:03.940923+00:00", "latency_seconds": 4.012119499966502, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "Arabic-script text. Words: \"واقعی\" (real, Urdu/Persian), \"رئیل\" (Real, transliteration), \"اسپورٹس\" (sports, with ٹ retroflex — Urdu-specific letter). Urdu.", "prediction": "ur", "usage": {"completion_tokens": 76, "prompt_tokens": 356, "total_tokens": 432, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 57, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010112, "usd": 0.0014944748051948053}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "189bb02cf486ae47907088220bc536c76c0bba561f4a9a25a1742d0b18d22dab"}, "error": null}
|
||||
{"id": "train-vi-000", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Một ứng cử viên alpha sẽ không cần sói chút nào cả .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87984cff86af330ad15f", "response_model": "kimi-k3", "response_created": 1785366425, "request_started_at": "2026-07-29T23:07:04.208168+00:00", "latency_seconds": 3.160106583032757, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text is Vietnamese: \"Một ứng cử viên alpha sẽ không cần sói chút nào cả.\" Diacritics and words are Vietnamese. Final label vi. Need final exact line only.", "prediction": "vi", "usage": {"completion_tokens": 72, "prompt_tokens": 371, "total_tokens": 443, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 53, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010012, "usd": 0.0014796955844155846}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "494da6aac1aed73d195822073c0687b3d87e999edaf613c431de503388e8d9bc"}, "error": null}
|
||||
{"id": "train-vi-001", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Cô ấy là một người phụ nữ đáng chú ý , tăng sức mạnh làm nhiếp chính cho những thanh niên tutmosis ii con trai của cô ấy trước khi lấy nó cho chính mình bằng cách tự xưng là quyền thống trị .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8799ad3e25d63f2c7976", "response_model": "kimi-k3", "response_created": 1785366425, "request_started_at": "2026-07-29T23:07:04.236735+00:00", "latency_seconds": 3.37666383292526, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text uses Vietnamese diacritics and vocabulary: \"Cô ấy là một người phụ nữ\" etc. Clearly vi. Need final exactly one line.", "prediction": "vi", "usage": {"completion_tokens": 59, "prompt_tokens": 447, "total_tokens": 506, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 40, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010232, "usd": 0.0015122098701298703}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "40747bcbb7e9db77be0aafde83a6df13a1d2f46aceaec52ed40a208b4957b522"}, "error": null}
|
||||
{"id": "train-vi-002", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Những sự kiện đặc biệt như thú cưng cho thấy , thử thách chó , và những ngày thái lịch sử được tổ chức trong suốt mùa hề .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879901c438c1be93d289", "response_model": "kimi-k3", "response_created": 1785366426, "request_started_at": "2026-07-29T23:07:05.070117+00:00", "latency_seconds": 4.363648499827832, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text has Vietnamese diacritics: Những, sự kiện, đặc biệt, thú cưng, thử thách, chó, ngày thái, lịch sử, được tổ chức, trong suốt, mùa hè. This is clearly Vietnamese. Final Answer: vi.", "prediction": "vi", "usage": {"completion_tokens": 103, "prompt_tokens": 410, "total_tokens": 513, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 84, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013892, "usd": 0.002053129350649351}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "9a2b1486bf2227bc1b47ef40f8f4a2c4e61899063dc0e70555916616bc461467"}, "error": null}
|
||||
{"id": "train-vi-003", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Bây giờ hãy nhìn vào số tiền mà chính phủ có thể tiết kiệm nếu họ không có tất cả những ngày nghỉ trong những ngày nghỉ đó .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a8799a5f8e9319959d65a", "response_model": "kimi-k3", "response_created": 1785366426, "request_started_at": "2026-07-29T23:07:05.071744+00:00", "latency_seconds": 3.2689064159058034, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text is Vietnamese with diacritics and vocabulary: \"Bây giờ hãy nhìn vào số tiền...\" Clearly vi. Need final exactly format.", "prediction": "vi", "usage": {"completion_tokens": 61, "prompt_tokens": 413, "total_tokens": 474, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 42, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009752, "usd": 0.0014412696103896106}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c1d2e40d8f8e8cf4573c5ee327c9c54ba060e73e7f151cc1f98fee4de9726e4c"}, "error": null}
|
||||
{"id": "train-vi-004", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "The Minato Mirai 21 dự án , được khởi chạy vào giữa những năm 1980 , đã được dự định sẽ biến một đường rộng lớn của bờ sông phía bắc và phía đông của sakuragi-cho vào một thành phố mô hình của tương lai , tích hợp kinh doanh , triển lãm , và giải trí cơ sở .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879b166409dfd708b61a", "response_model": "kimi-k3", "response_created": 1785366428, "request_started_at": "2026-07-29T23:07:06.514946+00:00", "latency_seconds": 6.2963415826670825, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text contains Vietnamese diacritics and vocabulary: \"dự án\", \"được khởi chạy\", \"vào giữa những năm\", \"thành phố\", \"tương lai\", \"kinh doanh\", \"triển lãm\", \"giải trí\". It's Vietnamese with some English proper nouns (Minato Mirai 21). Classify vi.", "prediction": "vi", "usage": {"completion_tokens": 118, "prompt_tokens": 477, "total_tokens": 595, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 99, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016732, "usd": 0.0024728592207792213}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "bfefb9549253fd321bc63ad20a0a4a9d9f40a3f0a1f4d08f353e3fead6676b58"}, "error": null}
|
||||
{"id": "train-vi-005", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Tiêu chuẩn quốc tế một số công ty lớn / công ty e đã bảo vệ 9000 chứng nhận là một tổ chức chất lượng ( ISO ) 9000 hoạt động kiểm soát .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879bfd4fe8c2d385a2d2", "response_model": "kimi-k3", "response_created": 1785366428, "request_started_at": "2026-07-29T23:07:06.614705+00:00", "latency_seconds": 4.380506957881153, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text contains Vietnamese diacritics and vocabulary: \"Tiêu chuẩn quốc tế\", \"một số\", \"công ty\", \"chứng nhận\", \"tổ chức\", \"chất lượng\", \"hoạt động kiểm soát\". Clear Vietnamese. Final vi.", "prediction": "vi", "usage": {"completion_tokens": 99, "prompt_tokens": 416, "total_tokens": 515, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 80, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013612, "usd": 0.0020117475324675327}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e3d1f7ffbc7324af5fc8aeb89fdc89bbc081524c5d7f08155bf8ffffcddfaa8a"}, "error": null}
|
||||
{"id": "train-vi-006", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Tất cả các bạn sẽ đến ăn tối với tôi ở savoy .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879b166409dfd708b61c", "response_model": "kimi-k3", "response_created": 1785366429, "request_started_at": "2026-07-29T23:07:07.369104+00:00", "latency_seconds": 2.822052333969623, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "Vietnamese sentence.", "prediction": "vi", "usage": {"completion_tokens": 25, "prompt_tokens": 374, "total_tokens": 399, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 6, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005372, "usd": 0.0007939397402597403}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "bb1a696984a6eb87c8489b3a550a440f58bf5fb4cdde9c185872c5f5cf784f27"}, "error": null}
|
||||
{"id": "train-vi-007", "split": "train", "source_label": "vi", "gold_label": "vi", "text": "Ca ' daan ' không thể thấy được nếu những kỵ sĩ đen nói bất cứ điều gì để trở lại .", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879cf1ac6b519f479432", "response_model": "kimi-k3", "response_created": 1785366429, "request_started_at": "2026-07-29T23:07:07.614421+00:00", "latency_seconds": 4.769405041821301, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text contains Vietnamese diacritics and words: không thể thấy được nếu những kỵ sĩ đen nói bất cứ điều gì để trở lại. This is clearly Vietnamese despite odd spacing/apostrophes. Label vi. Need final exactly one line.", "prediction": "vi", "usage": {"completion_tokens": 94, "prompt_tokens": 391, "total_tokens": 485, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 75, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012612, "usd": 0.0018639553246753249}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "6719799048c8f29de2b4cefeee6c6eeef887c0f4cfe16be18d78b3f335f26b06"}, "error": null}
|
||||
{"id": "train-zh-000", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "东西不错，携带方便。但中国人鼻子比较低的，用的时候容易掉下来。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879dc3316ea8a34aa6ef", "response_model": "kimi-k3", "response_created": 1785366430, "request_started_at": "2026-07-29T23:07:07.953738+00:00", "latency_seconds": 3.8785180002450943, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text uses Chinese characters: 东西不错，携带方便。但中国人鼻子比较低的，用的时候容易掉下来。 This is clearly Chinese (Simplified). Need final exact line.", "prediction": "zh", "usage": {"completion_tokens": 57, "prompt_tokens": 361, "total_tokens": 418, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 38, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008312, "usd": 0.0012284488311688313}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "07273c95fcab56a7e3ee536df0c2bfb185bd0f8576e5e1db973d6944c285ae76"}, "error": null}
|
||||
{"id": "train-zh-001", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "商品预测还有两周送货，我就提前申请退货取消订单，但是还是送过来了，联系客服怎么处理？客服说海外购订单无法取消，呵呵，买了就不许退，就这么硬气，然后退货必须自己出运费，什么玩意，垃圾商家亚马逊", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879d407571c0f21de54e", "response_model": "kimi-k3", "response_created": 1785366430, "request_started_at": "2026-07-29T23:07:08.341420+00:00", "latency_seconds": 4.2515935827977955, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is entirely in Chinese characters (Simplified). Clear Chinese signal with words like 商品, 预测, 送货, 退货, 客服, 亚马逊. Label zh.", "prediction": "zh", "usage": {"completion_tokens": 58, "prompt_tokens": 398, "total_tokens": 456, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 39, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009152, "usd": 0.001352594285714286}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0cc6909cd34fcc301f53a9a128df23465656600cbf47aba8fba71d4e9ffaf4d3"}, "error": null}
|
||||
{"id": "train-zh-002", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "本书共计368页，前57页包括解释全文，答记者问，理解适用，逐条释义，57页到165页案例，剩下部分都是法条，这么高定价，太黑了，请各位参考购买", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879db9c19ffa82a80169", "response_model": "kimi-k3", "response_created": 1785366429, "request_started_at": "2026-07-29T23:07:08.582521+00:00", "latency_seconds": 2.346033832989633, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is Chinese. Need final exactly format. Ensure no extra. Classify zh.", "prediction": "zh", "usage": {"completion_tokens": 38, "prompt_tokens": 389, "total_tokens": 427, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 19, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006972, "usd": 0.0010304072727272727}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "67bfa5cb351c8739ffd7bf09322b716ae2c243562a6ef2f6aef627abb7374292"}, "error": null}
|
||||
{"id": "train-zh-003", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "今天收到衣服， 很感人的卡片，真是体现的服务。超赞！", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879d4d654f2ee1729b56", "response_model": "kimi-k3", "response_created": 1785366429, "request_started_at": "2026-07-29T23:07:08.597435+00:00", "latency_seconds": 3.3248259578831494, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The user wants language classification. The text is Chinese: 今天收到衣服，很感人的卡片，真是体现的服务。超赞！ Native Chinese characters, clearly zh. Need output exactly one line Final Answer: zh. Ensure no extra. Need final exactly.", "prediction": "zh", "usage": {"completion_tokens": 72, "prompt_tokens": 360, "total_tokens": 432, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 53, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009792, "usd": 0.001447181298701299}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "277cc11c732013631efc78b71df1b3bd5a093c9b795c3a70409ecf2bbb2a07ba"}, "error": null}
|
||||
{"id": "train-zh-004", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "2016年4月11日下午4：31收到的天美时手表打开发现手表直接用表带绑在表盒内的塑料支架上，没有任何贴膜等保护措施，表带又轻又软质感很差，随后当天申请退货，4月12日上午顺丰快递寄回客服发的亚马逊钟表/珠宝频道，4月14日接到天津亚马逊打来电话，说经亚马逊检验手表没有贴膜且有磨损不符合退货要求。这就奇怪了 ，为什么寄出的时候根本不存在的贴膜，在退货检验的时候就变成理由了呢；为什么寄出的时候没有任何保护措施且直接绑在一个硬塑料支架上的这种包装方式反倒成了退货时手表有痕迹拒绝退货的理由了；而且网站的包装清单上没有注明任何关于这款手表应该有贴膜信息，如果说新品真的包含贴膜，那只能说明亚马逊将二手货和新品混合出售，这样作为消费者根本无法保护自己的权益；现在就造成亚马逊随便找一个理由就可以把这种二手产品直接推给消费者，这难道不属于变相的欺诈么。在此提醒其他想要购买亚马逊产品的消费者，至少中国亚马逊是没有道德底线的，高价商品千万不能在这购买。4月14日将这个情况反映到亚马逊之后，客服说会备注信息向上反映，4月16日再联系亚马逊时，被告知没有任何人员跟踪此事，亚马逊又开始采取拖延战术了。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879da95a610865f1e393", "response_model": "kimi-k3", "response_created": 1785366430, "request_started_at": "2026-07-29T23:07:08.884373+00:00", "latency_seconds": 2.9418548750691116, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is entirely in Chinese characters, describing a complaint about Amazon China and a watch return. Native script is Chinese (Simplified). Label zh. Final only.", "prediction": "zh", "usage": {"completion_tokens": 53, "prompt_tokens": 602, "total_tokens": 655, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 34, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012732, "usd": 0.0018816903896103898}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8ed3448b0b81a1a9553f0ca96284aa8173103ba048f2311a41b8ddff9b6671f1"}, "error": null}
|
||||
{"id": "train-zh-005", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "清仓的东西包装就这样 盒子发来就是开的还不是我打开的。不过东西不错啊 应该是全新没问题", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879efd361b86e4229935", "response_model": "kimi-k3", "response_created": 1785366430, "request_started_at": "2026-07-29T23:07:09.434482+00:00", "latency_seconds": 3.410837583243847, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is in Chinese characters. Phrases like 清仓的东西包装就这样 盒子发来就是开的... clearly Simplified Chinese. Label zh. Final only line.", "prediction": "zh", "usage": {"completion_tokens": 53, "prompt_tokens": 366, "total_tokens": 419, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 34, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008012, "usd": 0.001184111168831169}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a62976208275e26a0a840d04a0f6da9ca9099e3488c90e0a383b2ab2dada8bd9"}, "error": null}
|
||||
{"id": "train-zh-006", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "是我在亚马逊够买最不满意的物品，为什么换了这型号芯，出来的水烧开后有水渍，反而自来水却没有。为什么？厂家能解释吗？", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879e908217277579889a", "response_model": "kimi-k3", "response_created": 1785366431, "request_started_at": "2026-07-29T23:07:09.868394+00:00", "latency_seconds": 3.0820752498693764, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The user wants classification only. Need final exactly one line. Text is Chinese (Simplified), despite a typo 够买. Classify zh. Need final only. Ensure visible response exactly format.", "prediction": "zh", "usage": {"completion_tokens": 60, "prompt_tokens": 375, "total_tokens": 435, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 41, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008892, "usd": 0.001314168311688312}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "69cd737b60eb95d612f640f42fd2d885cdb40722dd62e9cf6c4f6fba931409cd"}, "error": null}
|
||||
{"id": "train-zh-007", "split": "train", "source_label": "zh", "gold_label": "zh", "text": "这车的价格太贵了，以为这个价格可以买好点的玩具，一打开包装我也是醉了，有种被骗的心", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879f8daee05e748dc32b", "response_model": "kimi-k3", "response_created": 1785366432, "request_started_at": "2026-07-29T23:07:10.191978+00:00", "latency_seconds": 3.5484760417602956, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is clearly Chinese (Simplified): 这车的价格太贵了... classify zh. Need output exactly one line Final Answer: zh. Ensure no extra. Final only.", "prediction": "zh", "usage": {"completion_tokens": 56, "prompt_tokens": 367, "total_tokens": 423, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 37, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008332, "usd": 0.0012314046753246755}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "731feca22761ef65e0ef9c61912e1633a3f70005e6ff69c7980fb5e24ac06645"}, "error": null}
|
||||
{"id": "test-ar-000", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "إن الاستمرارية ، بعد كل شيء ، هي فضيلة ، أو هكذا يقول أولئك الذين لم يفرضوها عليهم.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879f5e8ca4ca938894a4", "response_model": "kimi-k3", "response_created": 1785366432, "request_started_at": "2026-07-29T23:07:10.324159+00:00", "latency_seconds": 4.40911754174158, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is Arabic script: \"إن الاستمرارية ، بعد كل شيء ، هي فضيلة ، أو هكذا يقول أولئك الذين لم يفرضوها عليهم.\" Uses Arabic letters, common Arabic words إن، بعد كل شيء، هي، فضيلة، الذين. No Urdu-specific letters. Label ar. Need final exact line only.", "prediction": "ar", "usage": {"completion_tokens": 109, "prompt_tokens": 385, "total_tokens": 494, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 90, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013992, "usd": 0.0020679085714285714}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "008d4bee65fbf8d02c656ed585f39dab98e4355ab1188c7b17ac4bc4d950cab8"}, "error": null}
|
||||
{"id": "test-ar-001", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "حاولت مكاتب السياحة إعادة تسمية منطقة L'Estrie ، ولكن حتى أكثر الكيوبيين المتشددين تشدد على العرف ، إذا كان تقريبًا ، ترجمة للمقاطعات.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a879f9eab2b3298a1c583", "response_model": "kimi-k3", "response_created": 1785366432, "request_started_at": "2026-07-29T23:07:10.929211+00:00", "latency_seconds": 3.7444961671717465, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is mostly Arabic script. It contains Arabic words like \"حاولت مكاتب السياحة إعادة تسمية منطقة\" and includes a French proper noun L'Estrie. The dominant language is Arabic. Ignore URLs etc. Proper noun doesn't change. Final ar.", "prediction": "ar", "usage": {"completion_tokens": 82, "prompt_tokens": 408, "total_tokens": 490, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 63, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011752, "usd": 0.0017368540259740262}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "9b5cd00cc4c45494ad95b38414e10ae155cb37d5a829a24e9913dac6913aedc5"}, "error": null}
|
||||
{"id": "test-ar-002", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "حاولت مكاتب السياحة إعادة تسمية منطقة L'Estrie ، ولكن حتى أكثر الكيوبيين المتشددين تشدد على العرف ، إذا كان تقريبًا ، ترجمة للمقاطعات.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a05eaf930e38bfb420", "response_model": "kimi-k3", "response_created": 1785366433, "request_started_at": "2026-07-29T23:07:10.996053+00:00", "latency_seconds": 5.6210457920096815, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is in Arabic script with Arabic vocabulary: حاولت مكاتب السياحة إعادة تسمية منطقة ... ولكن حتى أكثر الكيوبيين المتشددين... Contains a French proper noun L'Estrie but rule says ignore proper nouns / don't decide from one proper noun. The dominant language is Arabic. No Urdu-specific letters. Label ar. Need output exact final line only.", "prediction": "ar", "usage": {"completion_tokens": 113, "prompt_tokens": 408, "total_tokens": 521, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 94, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014852, "usd": 0.0021950098701298704}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "9b5cd00cc4c45494ad95b38414e10ae155cb37d5a829a24e9913dac6913aedc5"}, "error": null}
|
||||
{"id": "test-ar-003", "split": "test", "source_label": "ar", "gold_label": "ar", "text": "هذا هو رئيس الرقيب كليم فرانسيس ، متقاعد من القوات الجوية الأمريكية.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a0e3dfcfab6fd101b8", "response_model": "kimi-k3", "response_created": 1785366433, "request_started_at": "2026-07-29T23:07:11.826980+00:00", "latency_seconds": 4.147148500196636, "attempt": 0, "content": "Final Answer: ar", "reasoning_content": "The text is Arabic script. It says: \"This is Sergeant Chief Clem Francis, retired from the US Air Force.\" Contains Arabic words هذا هو رئيس الرقيب ... متقاعد من القوات الجوية الأمريكية. Proper noun كليم فرانسيس is transliterated but overall Arabic. No Urdu-specific letters. Label ar. Need final exactly format.", "prediction": "ar", "usage": {"completion_tokens": 98, "prompt_tokens": 373, "total_tokens": 471, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 79, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012652, "usd": 0.0018698670129870133}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "68a6b15d0c0f264a07330bdb42f19a10279cece20446f7002ff365badcb88100"}, "error": null}
|
||||
{"id": "test-bg-000", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "Освен че проучи внимателно различни официални документи, чешкото правителство разгледа също така снимки от охранителни камери, направени пред посолството на Ирак.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a0567e9c2404b6c509", "response_model": "kimi-k3", "response_created": 1785366433, "request_started_at": "2026-07-29T23:07:11.832822+00:00", "latency_seconds": 6.459862458985299, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Cyrillic. Words like \"Освен че\", \"проучи внимателно\", \"различни официални документи\", \"чешкото правителство\", \"разгледа също така\", \"снимки\", \"охранителни камери\", \"направени пред посолството на Ирак\". This is Bulgarian, which is not in the listed labels (ru is Russian but this is Bulgarian: \"също така\", \"чешкото\", definite article -то, \"пред посолството\"). Not Russian. Label ot.", "prediction": "ot", "usage": {"completion_tokens": 162, "prompt_tokens": 412, "total_tokens": 574, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 143, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.019832, "usd": 0.002931015064935065}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "66a249281b2f31e31aef750b6752f0979e19b8d40e0c41bd77ae492344d18b1e"}, "error": null}
|
||||
{"id": "test-bg-001", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "Общата биология всъщност е на една ръка разстояние.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a0f4fe3c88d68d692e", "response_model": "kimi-k3", "response_created": 1785366433, "request_started_at": "2026-07-29T23:07:11.922868+00:00", "latency_seconds": 6.333595541771501, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Cyrillic. \"Общата биология всъщност е на една ръка разстояние.\" Words: общата (the general/common), биология (biology), всъщност (actually), е (is), на една ръка разстояние (at arm's length). This looks Bulgarian, not Russian. Russian would be \"Общая биология на самом деле на расстоянии вытянутой руки\". Bulgarian-specific: всъщност, ръка (with ъ), една. Bulgarian is not in labels; Cyrillic but not ru => ot. Need final exact line.", "prediction": "ot", "usage": {"completion_tokens": 165, "prompt_tokens": 366, "total_tokens": 531, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 146, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.019212, "usd": 0.0028393838961038964}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "881c5720694dc1f236cdf5d93265aaed19b044c97480dad670124647e176d868"}, "error": null}
|
||||
{"id": "test-bg-002", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "Тънкостите на галското подчинително наклонение изобщо не го притесняват и той дори не се безпокои, когато се упражнява.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a0697b682b780943aa", "response_model": "kimi-k3", "response_created": 1785366433, "request_started_at": "2026-07-29T23:07:12.384576+00:00", "latency_seconds": 4.732099291868508, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Cyrillic. It appears to be Bulgarian, not Russian (no ё, uses ъ and Bulgarian vocabulary/grammar like \"Тънкостите\", \"подчинително наклонение\", \"притесняват\", \"упражнява\"). Bulgarian is outside the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 100, "prompt_tokens": 395, "total_tokens": 495, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 81, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013292, "usd": 0.0019644540259740263}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "75f52cb030051d813f050f39dae44cb771aa7dd482f642f5a49432f6b31850a4"}, "error": null}
|
||||
{"id": "test-bg-003", "split": "test", "source_label": "bg", "gold_label": "ot", "text": "80% от участниците ще покажат повишени умения за разрешаване на конфликти.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a1fabdd4e38cb710f4", "response_model": "kimi-k3", "response_created": 1785366433, "request_started_at": "2026-07-29T23:07:12.593967+00:00", "latency_seconds": 3.5986146251671016, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Bulgarian: \"80% от участниците ще покажат повишени умения за разрешаване на конфликти.\" Cyrillic but Bulgarian, not Russian (ща, участниците, конфликти). Bulgarian not in labels, so ot. Need output exactly Final Answer: ot.", "prediction": "ot", "usage": {"completion_tokens": 89, "prompt_tokens": 372, "total_tokens": 461, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 70, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011732, "usd": 0.0017338981818181818}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a8a377f03e12bc87fa152a154b18376edc4e44d64c0e240861fb5b7fa5e1c414"}, "error": null}
|
||||
{"id": "test-de-000", "split": "test", "source_label": "de", "gold_label": "de", "text": "Die Abdeckung (dünnes Plastik)war leider gebrochen! Habe Ersatz angefordert, hat auch super geklappt, nur das die Lampe wieder den selben Schaden hat... wenn mich nicht alles täuscht wurde mir die gleiche einfach nochmal geschickt!!!!😈 Hoffentlich passts beim dritten mal!!!😡 Ansonsten gäbe es bei dem Preis eh nichts auszusetzen! Aber mehr Sterne hat sich der Service nicht verdient...", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a1f9a7d1afa0cb39bd", "response_model": "kimi-k3", "response_created": 1785366434, "request_started_at": "2026-07-29T23:07:12.812311+00:00", "latency_seconds": 3.3964453754015267, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text is clearly German: \"Die Abdeckung\", \"war leider gebrochen\", \"Habe Ersatz angefordert\", etc. Label de. Need output exactly one line.", "prediction": "de", "usage": {"completion_tokens": 61, "prompt_tokens": 460, "total_tokens": 521, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 42, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010692, "usd": 0.0015801942857142858}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a9e130df4d148abc93899eb5bf9d604faec7fe6a5ad04cdd64437a16cebef844"}, "error": null}
|
||||
{"id": "test-de-001", "split": "test", "source_label": "de", "gold_label": "de", "text": "Besser etwas größer bestellen !", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a12b9b250871ee024a", "response_model": "kimi-k3", "response_created": 1785366434, "request_started_at": "2026-07-29T23:07:12.845959+00:00", "latency_seconds": 3.520395875442773, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text \"Besser etwas größer bestellen !\" is German: Besser (better), etwas (somewhat), größer (larger with ß), bestellen (order). Label de. Need final exactly one line.", "prediction": "de", "usage": {"completion_tokens": 75, "prompt_tokens": 356, "total_tokens": 431, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 56, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010012, "usd": 0.0014796955844155846}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c0c8bd250bf39374b9745da9281633891e7266fb6aeb02984aa03edbadade283"}, "error": null}
|
||||
{"id": "test-de-002", "split": "test", "source_label": "de", "gold_label": "de", "text": "Lichter sind gut. Farben sind OK. Aber es hängt und es Klackert.--.Schade.--!!", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a1406b05b8c0525370", "response_model": "kimi-k3", "response_created": 1785366435, "request_started_at": "2026-07-29T23:07:12.951403+00:00", "latency_seconds": 5.623995791655034, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The user wants me to classify the language of the provided text. The text is: \"Lichter sind gut. Farben sind OK. Aber es hängt und es Klackert.--.Schade.--!!\"\n\nThis is German: \"Lichter sind gut\" (Lights are good), \"Farben sind OK\" (Colors are OK), \"Aber es hängt\" (But it hangs), \"Schade\" (Too bad). Clearly German.\n\nThe answer should be exactly: Final Answer: de", "prediction": "de", "usage": {"completion_tokens": 129, "prompt_tokens": 374, "total_tokens": 503, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 110, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015772, "usd": 0.002330978701298702}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "fd36ec79800439158a9a92df0c003294df5eb5cec6f3aadf0298c2f3d2ed411c"}, "error": null}
|
||||
{"id": "test-de-003", "split": "test", "source_label": "de", "gold_label": "de", "text": "Dicht und Robust, alles was er soll.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a33f9d8c7355c36059", "response_model": "kimi-k3", "response_created": 1785366436, "request_started_at": "2026-07-29T23:07:13.741185+00:00", "latency_seconds": 4.650652542244643, "attempt": 0, "content": "Final Answer: de", "reasoning_content": "The text \"Dicht und Robust, alles was er soll.\" contains German function words: und, alles, was, er, soll. Vocabulary Dicht/Robust. Clearly German. Need output exactly final line. Ensure no extra. Final Answer: de.", "prediction": "de", "usage": {"completion_tokens": 76, "prompt_tokens": 355, "total_tokens": 431, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 57, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010092, "usd": 0.0014915189610389611}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "14207a479cf4d925c904ca57b2ae3a1958a17b29866b248ded927e0b9ffd228f"}, "error": null}
|
||||
{"id": "test-el-000", "split": "test", "source_label": "el", "gold_label": "el", "text": "Ήταν η μοναδική απώλεια στην Κρίση της Κούβας και, ο Kaiser, πήρε τις φωτογραφίες και πέταξε κατευθείαν στο Αεροδρόμιο Andrews της Πολεμικής Αεροπορίας στην Ουάσινγκτον.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a39a50869de3b0e4fc", "response_model": "kimi-k3", "response_created": 1785366436, "request_started_at": "2026-07-29T23:07:14.674356+00:00", "latency_seconds": 5.431706874631345, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is in Greek script. It contains Greek words like Ήταν, η, μοναδική, απώλεια, στην, Κρίση, της, Κούβας, και, πήρε, τις, φωτογραφίες, etc. The proper nouns Kaiser, Andrews are Latin but embedded in Greek sentence. Native script is clearly Greek. Final Answer: el.", "prediction": "el", "usage": {"completion_tokens": 109, "prompt_tokens": 432, "total_tokens": 541, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 90, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014932, "usd": 0.0022068332467532472}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3e7e4032b32e1d2fc827a567435feafb8eb00dfe9442a8c7e5dd6144e8b16105"}, "error": null}
|
||||
{"id": "test-el-001", "split": "test", "source_label": "el", "gold_label": "el", "text": "Όμως, καθώς μεγάλωνε, ε, δεν αναγνώρισε ποτέ ότι έκανε λάθος, αλλά άλλαξε τη συμπεριφορά της.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a39a50869de3b0e4fd", "response_model": "kimi-k3", "response_created": 1785366436, "request_started_at": "2026-07-29T23:07:14.733959+00:00", "latency_seconds": 3.562251083087176, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "Greek text clearly: Όμως, καθώς... Greek script and vocabulary. Label el. Final exactly.", "prediction": "el", "usage": {"completion_tokens": 46, "prompt_tokens": 404, "total_tokens": 450, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 27, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008072, "usd": 0.0011929787012987013}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e714a858be2309b65894355ebcf64eb539cb0fdd6f6b05ef95aa07a7910136a5"}, "error": null}
|
||||
{"id": "test-el-002", "split": "test", "source_label": "el", "gold_label": "el", "text": "Και είναι διαφορετικό, όπως κάτω από κάθε πελάτη, είναι όλα τα αρχεία τους.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a45a513bb219e713b3", "response_model": "kimi-k3", "response_created": 1785366437, "request_started_at": "2026-07-29T23:07:15.974879+00:00", "latency_seconds": 3.4302475838921964, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is in Greek script: \"Και είναι διαφορετικό, όπως κάτω από κάθε πελάτη, είναι όλα τα αρχεία τους.\" Clearly Greek. Label el. Need output exactly Final Answer: el.", "prediction": "el", "usage": {"completion_tokens": 81, "prompt_tokens": 383, "total_tokens": 464, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 62, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011152, "usd": 0.0016481787012987015}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5cfb6903c275a5b02dcfc24f64b651b0eaca92c4a74c3520bb6c6ff6f7535423"}, "error": null}
|
||||
{"id": "test-el-003", "split": "test", "source_label": "el", "gold_label": "el", "text": "Και είναι διαφορετικό, όπως κάτω από κάθε πελάτη, είναι όλα τα αρχεία τους.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a50682347c2d84f9fa", "response_model": "kimi-k3", "response_created": 1785366438, "request_started_at": "2026-07-29T23:07:16.193210+00:00", "latency_seconds": 4.282676124945283, "attempt": 0, "content": "Final Answer: el", "reasoning_content": "The text is in Greek script and vocabulary: \"Και είναι διαφορετικό, όπως κάτω από κάθε πελάτη, είναι όλα τα αρχεία τους.\" Clearly Greek. Need final exactly one line.", "prediction": "el", "usage": {"completion_tokens": 78, "prompt_tokens": 383, "total_tokens": 461, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 59, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010852, "usd": 0.0016038410389610392}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5cfb6903c275a5b02dcfc24f64b651b0eaca92c4a74c3520bb6c6ff6f7535423"}, "error": null}
|
||||
{"id": "test-en-000", "split": "test", "source_label": "en", "gold_label": "en", "text": "We like the rack. There were some sharp pointed wires that needed to be cut before installing it.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a6ff812f22599fc03b", "response_model": "kimi-k3", "response_created": 1785366438, "request_started_at": "2026-07-29T23:07:16.209408+00:00", "latency_seconds": 3.7909603747539222, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The user wants language classification. Text is English. Need final exactly one line: Final Answer: en. Must not include reasoning. Ensure format.", "prediction": "en", "usage": {"completion_tokens": 49, "prompt_tokens": 364, "total_tokens": 413, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 30, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007572, "usd": 0.0011190825974025974}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8ec0b049d013c2e8c232a73b84719e73feae6f240f9599325fb2a0d9639d6731"}, "error": null}
|
||||
{"id": "test-en-001", "split": "test", "source_label": "en", "gold_label": "en", "text": "Love the gloves , but be,careful if you are allergic to nickel don't buy them they are made with nickel chloride which I am allergic very highly allergic to so that was the only downfall", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a4f533e66c8281cafd", "response_model": "kimi-k3", "response_created": 1785366437, "request_started_at": "2026-07-29T23:07:16.367068+00:00", "latency_seconds": 2.6454139999113977, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "The text is in English: \"Love the gloves, but be careful if you are allergic to nickel...\" Clearly English vocabulary and grammar. Output en.", "prediction": "en", "usage": {"completion_tokens": 50, "prompt_tokens": 384, "total_tokens": 434, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 31, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008072, "usd": 0.0011929787012987013}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "98f318f06ae372ab01a7c5e68f43d9d9e0e0f472c0b8765b4cab189f9e0ca697"}, "error": null}
|
||||
{"id": "test-en-002", "split": "test", "source_label": "en", "gold_label": "en", "text": "Ordered this for my son. We waited for 3weeks to arrive. The left side where the whole for the garter is torn so we had to tape it. The black paint was erased after a week. This mask is really thin.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a5a5c09dd52de39349", "response_model": "kimi-k3", "response_created": 1785366437, "request_started_at": "2026-07-29T23:07:16.617799+00:00", "latency_seconds": 1.8805179167538881, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "English product review.", "prediction": "en", "usage": {"completion_tokens": 24, "prompt_tokens": 394, "total_tokens": 418, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 5, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005672, "usd": 0.0008382774025974027}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "bf61f9511061c498d9b4174b423c80e6cb8b68b83e7779b24ee57ee219402e36"}, "error": null}
|
||||
{"id": "test-en-003", "split": "test", "source_label": "en", "gold_label": "en", "text": "Bought a few for craft show tables. I’ve owned them for years & I just can’t get the wrinkles out.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a531fa7ecf69e5f80f", "response_model": "kimi-k3", "response_created": 1785366438, "request_started_at": "2026-07-29T23:07:17.117317+00:00", "latency_seconds": 1.8313229577615857, "attempt": 0, "content": "Final Answer: en", "reasoning_content": "English text.", "prediction": "en", "usage": {"completion_tokens": 23, "prompt_tokens": 369, "total_tokens": 392, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 4, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005072, "usd": 0.000749602077922078}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "a67e4185b00cb81d4fef29714be9ee2ec14e92e4259ca063e01d3081a1b17381"}, "error": null}
|
||||
{"id": "test-es-000", "split": "test", "source_label": "es", "gold_label": "es", "text": "La compré para mi mujer y le ha gustado bastante. Es muy grande, pero dormimos en una cama de 1,50m y no hemos tenido problema de espacio. La única pega que encuentra es que es baja para dormir sin almohada, si pone la almohada debajo le resulta demasiado alta. También la usa en el sofá. Los materiales son de calidad y agradables", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a7776710264cf67d00", "response_model": "kimi-k3", "response_created": 1785366439, "request_started_at": "2026-07-29T23:07:18.257181+00:00", "latency_seconds": 4.291270665824413, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The user wants me to classify the language of the provided text. The text is clearly Spanish: \"La compré para mi mujer y le ha gustado bastante...\" contains Spanish vocabulary, accents, function words like \"para\", \"mi\", \"y\", \"le\", \"ha\", \"es\", etc. The label should be es.\n\nI need to output exactly: Final Answer: es", "prediction": "es", "usage": {"completion_tokens": 100, "prompt_tokens": 449, "total_tokens": 549, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 81, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014372, "usd": 0.0021240696103896105}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8842deca6a97ab50e5b0551ea999723eaaa91cda436de746297ae8abdc7eb68c"}, "error": null}
|
||||
{"id": "test-es-001", "split": "test", "source_label": "es", "gold_label": "es", "text": "No se sujeta bien al secador y es un secador universal con boquilla larga. No es satisfactorio para mi agrado", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a7776710264cf67d01", "response_model": "kimi-k3", "response_created": 1785366440, "request_started_at": "2026-07-29T23:07:18.293434+00:00", "latency_seconds": 4.277430208865553, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The text is Spanish: \"No se sujeta bien al secador y es un secador universal con boquilla larga. No es satisfactorio para mi agrado\". Clear Spanish vocabulary and grammar. Need output exactly Final Answer: es.", "prediction": "es", "usage": {"completion_tokens": 72, "prompt_tokens": 376, "total_tokens": 448, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 53, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010112, "usd": 0.0014944748051948053}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "2613cf5d66f1e0c79272442f34bc71ff12bb1deee8dc4d711094e848f5e13e6a"}, "error": null}
|
||||
{"id": "test-es-002", "split": "test", "source_label": "es", "gold_label": "es", "text": "Han tardado el tiempo que decía en la entrega, y todo estaba bien. Pero creo que le falta una guía/soporte, al igual que otros protectores de pantalla tienen para ser más sencillos fe colocar. También creo que no es exactamente igual que el OPPO A3, pero aún así ha quedado bien.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a6fd4fe8c2d385a2e1", "response_model": "kimi-k3", "response_created": 1785366439, "request_started_at": "2026-07-29T23:07:18.296826+00:00", "latency_seconds": 3.1743119577877223, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The text is Spanish: \"Han tardado el tiempo que decía en la entrega...\" clear Spanish vocabulary and grammar. Need final exactly. Ensure no extra. Final Answer: es.", "prediction": "es", "usage": {"completion_tokens": 58, "prompt_tokens": 428, "total_tokens": 486, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 39, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009752, "usd": 0.0014412696103896106}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b3bf8f80acb921955b52a0ff920bcd693746bed8f2810a5a17af7280eb821502"}, "error": null}
|
||||
{"id": "test-es-003", "split": "test", "source_label": "es", "gold_label": "es", "text": "Buenas tardes. No me ha llegado el pedido del libro y pone que está entregado. Espero su contestación lo antes posible gracias", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a655543c615b081379", "response_model": "kimi-k3", "response_created": 1785366439, "request_started_at": "2026-07-29T23:07:18.392705+00:00", "latency_seconds": 3.3748448751866817, "attempt": 0, "content": "Final Answer: es", "reasoning_content": "The user wants me to classify the language. The text is clearly Spanish: \"Buenas tardes. No me ha llegado el pedido del libro y pone que está entregado. Espero su contestación lo antes posible gracias\". Spanish vocabulary and structure. Final answer es.", "prediction": "es", "usage": {"completion_tokens": 82, "prompt_tokens": 381, "total_tokens": 463, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 63, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011212, "usd": 0.001657046233766234}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "67037286cffab38fcff4919a3fd0a0f5c492919f02ef8073503b5de45bc8b5e8"}, "error": null}
|
||||
{"id": "test-fr-000", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "Au bout de 1 mois quasiment casser j'ai du re commander une coque. Le miroir est très fragile et le tissu sur le bord aussi", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a7fd361b86e422993e", "response_model": "kimi-k3", "response_created": 1785366439, "request_started_at": "2026-07-29T23:07:18.498914+00:00", "latency_seconds": 3.7699305000714958, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text is French: \"Au bout de 1 mois quasiment casser j'ai du re commander une coque. Le miroir est très fragile et le tissu sur le bord aussi\". Clear French function words: au bout de, j'ai, une, le, est très, et, sur, aussi. Label fr.", "prediction": "fr", "usage": {"completion_tokens": 95, "prompt_tokens": 384, "total_tokens": 479, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 76, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012572, "usd": 0.0018580436363636365}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f33f1a9d9d1f4bfd8346653f3866e72b147675d28a496a4932894b5b3cf0fd48"}, "error": null}
|
||||
{"id": "test-fr-001", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "livraison vraiment désastreux par dhl et colissimo voir les photos en dessous", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a70682347c2d84f9fd", "response_model": "kimi-k3", "response_created": 1785366439, "request_started_at": "2026-07-29T23:07:18.576129+00:00", "latency_seconds": 3.159727666992694, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text is French: \"livraison vraiment désastreux par dhl et colissimo voir les photos en dessous\" means delivery really disastrous by DHL and Colissimo see photos below. Label fr.", "prediction": "fr", "usage": {"completion_tokens": 69, "prompt_tokens": 369, "total_tokens": 438, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 50, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009672, "usd": 0.001429446233766234}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "de9c5fb6d58cc864748e57aef5bb26bb3d31493d4d327753b4f752126496c0cc"}, "error": null}
|
||||
{"id": "test-fr-002", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "Copie chinoise de l'original. Qualité de finition moyenne, j'ai même un doute sur la sécurité. Le dock ne fonctionne pas avec. pas assez puissant.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a75657fe29c59a3ab8", "response_model": "kimi-k3", "response_created": 1785366440, "request_started_at": "2026-07-29T23:07:18.949260+00:00", "latency_seconds": 3.511076582595706, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "The text is French: \"Copie chinoise de l'original. Qualité de finition moyenne, j'ai même un doute sur la sécurité...\" Clear French vocabulary and function words. Label fr. Need output exactly final answer.", "prediction": "fr", "usage": {"completion_tokens": 74, "prompt_tokens": 391, "total_tokens": 465, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 55, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010612, "usd": 0.0015683709090909092}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5d7a9ff910edadb5a659af3c68bd3739916a4078814744f2722cca235ae5656a"}, "error": null}
|
||||
{"id": "test-fr-003", "split": "test", "source_label": "fr", "gold_label": "fr", "text": "compliqué ce livre mais après chaque envoi je vous enverrai une satisfaction du moment que le transporteur est bpost plus facile", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a8cf84bf1e61caf0c0", "response_model": "kimi-k3", "response_created": 1785366441, "request_started_at": "2026-07-29T23:07:19.013067+00:00", "latency_seconds": 5.609529249835759, "attempt": 0, "content": "Final Answer: fr", "reasoning_content": "Text: \"compliqué ce livre mais après chaque envoi je vous enverrai une satisfaction du moment que le transporteur est bpost plus facile\". Words: compliqué, ce, livre, mais, après, chaque, envoi, je, vous, enverrai, une, satisfaction, du moment que, le transporteur est, plus facile. Clearly French despite ungrammatical. bpost proper noun Belgian post ignored. Label fr. Need final exact line.", "prediction": "fr", "usage": {"completion_tokens": 127, "prompt_tokens": 377, "total_tokens": 504, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 108, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015632, "usd": 0.0023102877922077923}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "729597b57e4dd3d4e771ffec4eb04b236346555739427d7edd1387e2abc503fd"}, "error": null}
|
||||
{"id": "test-hi-000", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "इसलिए मुझे नहीं पता कि मैंने तमन्ना की थी या नहीं |", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a75657fe29c59a3ab9", "response_model": "kimi-k3", "response_created": 1785366440, "request_started_at": "2026-07-29T23:07:19.405857+00:00", "latency_seconds": 3.292642083019018, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script with Hindi words: इसलिए मुझे नहीं पता कि मैंने तमन्ना की थी या नहीं. Clearly Hindi. Need output exactly Final Answer: hi.", "prediction": "hi", "usage": {"completion_tokens": 74, "prompt_tokens": 377, "total_tokens": 451, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 55, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010332, "usd": 0.0015269890909090909}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "77464b86eb62409449c4dc08cb47837579b8f1d8a4b80b3eebe4e7c7c8002b15"}, "error": null}
|
||||
{"id": "test-hi-001", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "इसलिए मुझे नहीं पता कि मैंने तमन्ना की थी या नहीं |", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a8fabdd4e38cb71102", "response_model": "kimi-k3", "response_created": 1785366441, "request_started_at": "2026-07-29T23:07:20.000994+00:00", "latency_seconds": 3.10679891705513, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script with Hindi words: इसलिए, मुझे, नहीं, पता, कि, मैंने. Clear Hindi. Final Answer: hi.", "prediction": "hi", "usage": {"completion_tokens": 64, "prompt_tokens": 377, "total_tokens": 441, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 45, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009332, "usd": 0.0013791968831168833}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "77464b86eb62409449c4dc08cb47837579b8f1d8a4b80b3eebe4e7c7c8002b15"}, "error": null}
|
||||
{"id": "test-hi-002", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "और, ज़ाहिर है, एंड्रोव ग्रोमिकोव ने कोई जवाब नहीं दिया, लेकिन हमारे पास यू 2 की फिल्मों के आधार पर सारी जानकारी थी।", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a900c578d2cbd45ceb", "response_model": "kimi-k3", "response_created": 1785366442, "request_started_at": "2026-07-29T23:07:20.106774+00:00", "latency_seconds": 5.213689208030701, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The user wants language classification. The text is in Devanagari script: \"और, ज़ाहिर है, एंड्रोव ग्रोमिकोव ने कोई जवाब नहीं दिया, लेकिन हमारे पास यू 2 की फिल्मों के आधार पर सारी जानकारी थी।\" This is Hindi. Need output exactly \"Final Answer: hi\". Ensure no extra. Done.", "prediction": "hi", "usage": {"completion_tokens": 124, "prompt_tokens": 412, "total_tokens": 536, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 105, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016032, "usd": 0.002369404675324676}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "2fa974fc36bc0bb5845ab07a776f7648360e2cfacc91a607c97d481458b46683"}, "error": null}
|
||||
{"id": "test-hi-003", "split": "test", "source_label": "hi", "gold_label": "hi", "text": "उह, मैं अभी भी एक और केवल नौ दो-दो था जिसने कभी रेगुलेटर पर इंजेक्शन सेट कर दिया था।", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87a91febc1a70bd565ef", "response_model": "kimi-k3", "response_created": 1785366442, "request_started_at": "2026-07-29T23:07:20.476567+00:00", "latency_seconds": 5.551016957964748, "attempt": 0, "content": "Final Answer: hi", "reasoning_content": "The text is in Devanagari script. It contains Hindi words like \"मैं\", \"अभी\", \"भी\", \"एक\", \"और\", \"केवल\", \"नौ\", \"था\", \"जिसने\", \"कभी\", \"पर\", \"सेट\", \"कर\", \"दिया\". Some transliterated English terms (रेगुलेटर, इंजेक्शन) appear but the sentence is clearly Hindi. Label hi.", "prediction": "hi", "usage": {"completion_tokens": 126, "prompt_tokens": 398, "total_tokens": 524, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 107, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.015952, "usd": 0.002357581298701299}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7b7f5fa39ae418065aaaf6c03b45e5ef8cc8857b2903b5348d126dd904f11d89"}, "error": null}
|
||||
{"id": "test-it-000", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Una persona sta affettando delle cipolle.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87aae7c97367e8040502", "response_model": "kimi-k3", "response_created": 1785366443, "request_started_at": "2026-07-29T23:07:21.471938+00:00", "latency_seconds": 2.8978641661815345, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Italian -> ot.", "prediction": "ot", "usage": {"completion_tokens": 24, "prompt_tokens": 356, "total_tokens": 380, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 5, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.004912, "usd": 0.0007259553246753247}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f4307924ae20be8f85189a45606bc8424a548c293b9a64ea6d849dbbb4d83d81"}, "error": null}
|
||||
{"id": "test-it-001", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Diversi bambini saltano su e giù su un trampolino.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87aa8dc9461ab3e2ffa3", "response_model": "kimi-k3", "response_created": 1785366443, "request_started_at": "2026-07-29T23:07:21.736498+00:00", "latency_seconds": 2.8002587920054793, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Italian not listed -> ot.", "prediction": "ot", "usage": {"completion_tokens": 26, "prompt_tokens": 362, "total_tokens": 388, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 7, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005232, "usd": 0.0007732488311688312}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "0ac54796bf717037742f5143e478a7403be3427f981237976ab87f9a8d4191ea"}, "error": null}
|
||||
{"id": "test-it-002", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Il tasso di disoccupazione civile è migliorato marginalmente il mese scorso, scendendo al 6,1 per cento, anche se le aziende hanno tagliato le buste paga di 93.000 unità.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87aa1cfe58969987281c", "response_model": "kimi-k3", "response_created": 1785366443, "request_started_at": "2026-07-29T23:07:21.767818+00:00", "latency_seconds": 4.7015627501532435, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Italian, which is not in the listed labels. Rule 4 says language outside listed set is ot. Need final exactly.", "prediction": "ot", "usage": {"completion_tokens": 48, "prompt_tokens": 403, "total_tokens": 451, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 29, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008252, "usd": 0.001219581298701299}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c39c7a52a0d0009fdf403665e0664126bd02972fae0289479558c6ec5b30c947"}, "error": null}
|
||||
{"id": "test-it-003", "split": "test", "source_label": "it", "gold_label": "ot", "text": "Myanmar arresta 44 persone per violenza settaria", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ab38085107f41a0e0d", "response_model": "kimi-k3", "response_created": 1785366444, "request_started_at": "2026-07-29T23:07:22.269201+00:00", "latency_seconds": 3.124396374914795, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Italian - \"arresta 44 persone per violenza\" → ot.", "prediction": "ot", "usage": {"completion_tokens": 37, "prompt_tokens": 358, "total_tokens": 395, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 18, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006252, "usd": 0.0009239968831168832}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "71991b779a3ebf0334c58708245adfed154f5596d5abe628fede97dce6ca98b8"}, "error": null}
|
||||
{"id": "test-ja-000", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "SSDへシステムの引っ越しに使いました。 簡単に目的は達成しました。 ソケットのバネが弱いようです。 簡単に外れてしまう ので・・・それなりの工夫が必要でした。 個体差はあるかもしれませんが・・・。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ab6aa29639cda94037", "response_model": "kimi-k3", "response_created": 1785366443, "request_started_at": "2026-07-29T23:07:22.460622+00:00", "latency_seconds": 3.345957832876593, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Japanese, which is not among the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh). According to rule 4, a language outside the listed set is ot. Final Answer: ot.", "prediction": "ot", "usage": {"completion_tokens": 78, "prompt_tokens": 424, "total_tokens": 502, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 59, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011672, "usd": 0.0017250306493506496}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3146b843da28ba39c62e0307b3e8de9c2b09c2a9268b614fe0c915f44c596c80"}, "error": null}
|
||||
{"id": "test-ja-001", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "軽いので持ち運びしやすいです。 車の中でも使えることも考慮すると手頃かなと思います。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87abc50aaca0e6feb921", "response_model": "kimi-k3", "response_created": 1785366444, "request_started_at": "2026-07-29T23:07:22.548731+00:00", "latency_seconds": 3.9219847079366446, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese: 軽いので持ち運びしやすいです。車の中でも使えることも考慮すると手頃かなと思います。 Japanese is not among listed labels, so classify as ot. Must output exactly Final Answer: ot.", "prediction": "ot", "usage": {"completion_tokens": 82, "prompt_tokens": 381, "total_tokens": 463, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 63, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011212, "usd": 0.001657046233766234}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b6d23bcba14c2d880648ff1161b01c368f82fb88dddc044d7ab5ab5628d5b253"}, "error": null}
|
||||
{"id": "test-ja-002", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "もう、見るべしとしか言えない。 でも、一部のヒロインが好きな方（僕）には、残念。出てこないんだもん。あんまり。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ab6439639453f16181", "response_model": "kimi-k3", "response_created": 1785366444, "request_started_at": "2026-07-29T23:07:22.571062+00:00", "latency_seconds": 3.094792208634317, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese, which is outside the listed language set, so label ot.", "prediction": "ot", "usage": {"completion_tokens": 37, "prompt_tokens": 392, "total_tokens": 429, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 18, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006932, "usd": 0.0010244955844155846}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "d2d6e6312f97c99f7b1f44f101cee2d17caf1f2cc43782f0220beb8fe1a2de43"}, "error": null}
|
||||
{"id": "test-ja-003", "split": "test", "source_label": "ja", "gold_label": "ot", "text": "なんだかちょいちょいバグるのか左に移動します。 が、繋ぎ直すと大丈夫！ でも、ちょっと面倒なので☆１つ減らしときました。", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87abbcb8b1ca0dc259f3", "response_model": "kimi-k3", "response_created": 1785366443, "request_started_at": "2026-07-29T23:07:22.699399+00:00", "latency_seconds": 3.32469749962911, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Japanese (hiragana/katakana/kanji: なんだかちょいちょいバグる...). Japanese is not among listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 68, "prompt_tokens": 398, "total_tokens": 466, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 49, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010152, "usd": 0.0015003864935064935}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "4efde5ba61738065fd71e51486eaaf8cb7a1861e4cdd915a5faad3f68dd804b1"}, "error": null}
|
||||
{"id": "test-nl-000", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "Het vlees, gevogelte, boter, kaas en noten werden een jaar geleden in beslag genomen in een LaGrou Cold Storage warehouse in Chicago.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87abe3dfcfab6fd101c1", "response_model": "kimi-k3", "response_created": 1785366444, "request_started_at": "2026-07-29T23:07:23.108448+00:00", "latency_seconds": 3.0263338750228286, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Dutch: \"Het vlees, gevogelte, boter, kaas en noten werden een jaar geleden in beslag genomen...\" Dutch is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 75, "prompt_tokens": 384, "total_tokens": 459, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 56, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010572, "usd": 0.0015624592207792208}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7c74dd7470f2b40c5407bfe37706f98a9d0f5e238ed6cc1444a2b58045437a96"}, "error": null}
|
||||
{"id": "test-nl-001", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "Een zwarte hond die naar de camera kijkt.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ac31fa7ecf69e5f816", "response_model": "kimi-k3", "response_created": 1785366445, "request_started_at": "2026-07-29T23:07:24.370528+00:00", "latency_seconds": 4.593882374931127, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user wants me to classify language of text. Need answer exactly Final Answer: xx. Need determine label. Text: \"Een zwarte hond die naar de camera kijkt.\" This is Dutch: A black dog looking at the camera. Dutch not listed among labels, so ot per rule 4 (outside listed set). Need final only. Ensure exact format. The system says visible response must be exactly Final Answer: xx. Need output Final Answer: ot. No extra.", "prediction": "ot", "usage": {"completion_tokens": 119, "prompt_tokens": 358, "total_tokens": 477, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 100, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014452, "usd": 0.0021358929870129873}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "aa3087871ed10651eb31c3ae771edf5c1d5b3180a11fbdb84c16f372e38bda0e"}, "error": null}
|
||||
{"id": "test-nl-002", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "Een schildpad zwemt in het water.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ad5eaf930e38bfb42a", "response_model": "kimi-k3", "response_created": 1785366445, "request_started_at": "2026-07-29T23:07:24.537453+00:00", "latency_seconds": 2.7004712093621492, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text \"Een schildpad zwemt in het water.\" is Dutch. Dutch is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 57, "prompt_tokens": 356, "total_tokens": 413, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 38, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008212, "usd": 0.0012136696103896105}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3d2290d2bad1caa0bdd1235d1442c771dcc0b470885eb07df8ceafb604f6cea6"}, "error": null}
|
||||
{"id": "test-nl-003", "split": "test", "source_label": "nl", "gold_label": "ot", "text": "De technologisch geplaatste Nasdaq Composite Index .IXIC daalde met 25,36 punten, of 1,53 procent, tot 1.628,26.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87adbcb8b1ca0dc259f8", "response_model": "kimi-k3", "response_created": 1785366445, "request_started_at": "2026-07-29T23:07:24.623547+00:00", "latency_seconds": 4.654026167001575, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Dutch: \"De technologisch geplaatste Nasdaq Composite Index daalde met punten, of procent, tot...\" Dutch is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh), so it should be classified as ot (other). Need visible exactly one line Final Answer: ot. Ensure no extra. ", "prediction": "ot", "usage": {"completion_tokens": 106, "prompt_tokens": 384, "total_tokens": 490, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 87, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013672, "usd": 0.002020615064935065}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "32c94f8abcbd006b9c507bd331e7bd48b83347c6006efb02514935af29edb050"}, "error": null}
|
||||
{"id": "test-pl-000", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Opalona dziewczyna w kwiatowym bikini pływa w ciemnoniebieskiej wodzie.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87aea5dea32f85472d79", "response_model": "kimi-k3", "response_created": 1785366447, "request_started_at": "2026-07-29T23:07:25.321404+00:00", "latency_seconds": 3.6388149582780898, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Polish: \"Opalona dziewczyna w kwiatowym bikini pływa w ciemnoniebieskiej wodzie.\" Polish is not in the listed labels, so classify as ot.", "prediction": "ot", "usage": {"completion_tokens": 67, "prompt_tokens": 372, "total_tokens": 439, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 48, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009532, "usd": 0.0014087553246753249}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "86c3b4bebbd836b4ebfe36512abefe1887a738d4d2f60ca0c6fe3b215542f558"}, "error": null}
|
||||
{"id": "test-pl-001", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Jeśli nie jesteś pewien, jak to zrobić, nie rób tego wcale.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ae220baf5a77ff2266", "response_model": "kimi-k3", "response_created": 1785366446, "request_started_at": "2026-07-29T23:07:25.394252+00:00", "latency_seconds": 3.997688375413418, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The user is asking to classify the language of a Polish sentence. The text is Polish: \"Jeśli nie jesteś pewien, jak to zrobić, nie rób tego wcale.\" Polish is not in the listed labels (ar, de, el, en, es, fr, hi, ru, tr, ur, vi, zh), so it falls under ot. Need final answer exactly one line: Final Answer: ot. Ensure no extra.", "prediction": "ot", "usage": {"completion_tokens": 119, "prompt_tokens": 370, "total_tokens": 489, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 100, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014692, "usd": 0.0021713631168831173}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f6458c13879066550fe4ff576e1d4de70fd36da9d36b46b2dc1cc165eb931ad7"}, "error": null}
|
||||
{"id": "test-pl-002", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Rada Kolebki Wolności nie jest pierwszą, która złamała stanowisko grupy narodowej.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ae8dc9461ab3e2ffa9", "response_model": "kimi-k3", "response_created": 1785366447, "request_started_at": "2026-07-29T23:07:25.666742+00:00", "latency_seconds": 2.4564355830661952, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Polish text, not in list → ot.", "prediction": "ot", "usage": {"completion_tokens": 30, "prompt_tokens": 374, "total_tokens": 404, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 11, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005872, "usd": 0.0008678358441558441}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f439e53fc642973c05aac843157ad6d5779dc646e9a36412459ca8199c7e12bd"}, "error": null}
|
||||
{"id": "test-pl-003", "split": "test", "source_label": "pl", "gold_label": "ot", "text": "Średnia dla przemysłu Dow Jones spadła o 10,89 dnia do poziomu 9 837,94, po awansie o 111,04 w środę.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ae31fa7ecf69e5f818", "response_model": "kimi-k3", "response_created": 1785366446, "request_started_at": "2026-07-29T23:07:25.807291+00:00", "latency_seconds": 1.9360032081604004, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Polish text, not in list → ot.", "prediction": "ot", "usage": {"completion_tokens": 30, "prompt_tokens": 390, "total_tokens": 420, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 11, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006192, "usd": 0.0009151293506493507}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "03ba5fa7bc3a94a955fd9fc868320c852fcf71005f3a205bbf8efe7e9f6c22bb"}, "error": null}
|
||||
{"id": "test-pt-000", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "A Kollar-Kotelly agendou outra audiência para Janeiro sobre o cumprimento do acordo antitrust.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87aec3316ea8a34aa703", "response_model": "kimi-k3", "response_created": 1785366447, "request_started_at": "2026-07-29T23:07:26.024822+00:00", "latency_seconds": 2.1977585833519697, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Portuguese, not in list → ot.", "prediction": "ot", "usage": {"completion_tokens": 29, "prompt_tokens": 370, "total_tokens": 399, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 10, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005692, "usd": 0.0008412332467532469}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "c27bd3398fbaf585ce36d4fb0fddcb0389e161c333117f7185bd8c6cc1c6bd42"}, "error": null}
|
||||
{"id": "test-pt-001", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "A jovem mulher está a namoriscar com o jovem rapaz.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87ae3d093ac098615567", "response_model": "kimi-k3", "response_created": 1785366447, "request_started_at": "2026-07-29T23:07:26.028271+00:00", "latency_seconds": 3.496589125134051, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text: \"A jovem mulher está a namoriscar com o jovem rapaz.\" This is Portuguese (\"jovem mulher\", \"está a\", \"namoriscar\", \"rapaz\"). Portuguese is not in the listed labels, so classify as ot. Final answer only.", "prediction": "ot", "usage": {"completion_tokens": 86, "prompt_tokens": 364, "total_tokens": 450, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 67, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011272, "usd": 0.0016659137662337665}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7e80f36c5d8f690df0605ff8b548d4edd5c08bb55d45aa098f267046084ce992"}, "error": null}
|
||||
{"id": "test-pt-002", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "Um grupo de homens joga futebol na praia.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87afb723fe59282dfdb6", "response_model": "kimi-k3", "response_created": 1785366447, "request_started_at": "2026-07-29T23:07:26.135361+00:00", "latency_seconds": 2.7028613328002393, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Portuguese not listed => ot.", "prediction": "ot", "usage": {"completion_tokens": 27, "prompt_tokens": 358, "total_tokens": 385, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 8, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.005252, "usd": 0.0007762046753246754}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "2ea099041385617aa57d4c4251a8501f0879798a9f9f1d7a1067a87366f183b9"}, "error": null}
|
||||
{"id": "test-pt-003", "split": "test", "source_label": "pt", "gold_label": "ot", "text": "Uma jovem com sujidade na cara e uma bicicleta vermelha de criança está no fundo.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764ddcf1695a654d8ae3", "response_model": "kimi-k3", "response_created": 1785689678, "request_started_at": "2026-08-02T16:54:35.477838+00:00", "latency_seconds": 4.7845044170098845, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Portuguese: \"Uma jovem com sujidade na cara e uma bicicleta vermelha de criança está no fundo.\" Portuguese is not in the listed labels, so classify as ot. Need final exact line only.", "prediction": "ot", "usage": {"completion_tokens": 73, "prompt_tokens": 372, "total_tokens": 445, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 54, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01474, "usd": 0.002178457142857143}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b2dbb14a773c33657e1f0503be905331dda9c7dd917104d50cff56f3e69ddec7"}, "error": null}
|
||||
{"id": "test-ru-000", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Моим детям сейчас двадцать один и двадцать четыре года, поэтому, мне не приходится", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6a87af2b9b250871ee0261", "response_model": "kimi-k3", "response_created": 1785366447, "request_started_at": "2026-07-29T23:07:26.472946+00:00", "latency_seconds": 3.5894932080991566, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The user wants classification only. Need output exactly 'Final Answer: xx'. Determine text Russian: Cyrillic, words Моим детям... Russian. Label ru. Ensure final only line.", "prediction": "ru", "usage": {"completion_tokens": 60, "prompt_tokens": 380, "total_tokens": 440, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 41, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008992, "usd": 0.0013289475324675325}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "167ebd9b2d70606366e8e263f18499e949ec95fbbe8ccd76a0d6d757144afe29"}, "error": null}
|
||||
{"id": "test-ru-001", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Но я торопился высадить тебя.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764dccfd8627e2f93862", "response_model": "kimi-k3", "response_created": 1785689677, "request_started_at": "2026-08-02T16:54:36.583891+00:00", "latency_seconds": 2.711229833017569, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is Cyrillic Russian: \"Но я торопился высадить тебя.\" Clearly Russian.", "prediction": "ru", "usage": {"completion_tokens": 44, "prompt_tokens": 357, "total_tokens": 401, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 25, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01154, "usd": 0.0017055220779220782}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7d68c3e0f7777946e57dd7bfe5eaae612c77c3b815ba72818a71dcc8e487ca0e"}, "error": null}
|
||||
{"id": "test-ru-002", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Но я торопился высадить тебя.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764ddcf1695a654d8ae4", "response_model": "kimi-k3", "response_created": 1785689678, "request_started_at": "2026-08-02T16:54:36.584229+00:00", "latency_seconds": 3.582240624993574, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is in Cyrillic script: \"Но я торопился высадить тебя.\" This is Russian, meaning \"But I was in a hurry to drop you off.\" Clear Russian vocabulary and grammar. Label: ru.", "prediction": "ru", "usage": {"completion_tokens": 69, "prompt_tokens": 357, "total_tokens": 426, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 50, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01404, "usd": 0.0020750025974025977}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "7d68c3e0f7777946e57dd7bfe5eaae612c77c3b815ba72818a71dcc8e487ca0e"}, "error": null}
|
||||
{"id": "test-ru-003", "split": "test", "source_label": "ru", "gold_label": "ru", "text": "Через каждые сто градусов пятна краски меняют свой цвет, она может быть красной и изменить цвет на синий.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764de802f55ebfc9747f", "response_model": "kimi-k3", "response_created": 1785689678, "request_started_at": "2026-08-02T16:54:36.584582+00:00", "latency_seconds": 4.101252792024752, "attempt": 0, "content": "Final Answer: ru", "reasoning_content": "The text is in Cyrillic script with Russian vocabulary and grammar: \"Через каждые сто градусов пятна краски меняют свой цвет...\" Clearly Russian. Label ru.", "prediction": "ru", "usage": {"completion_tokens": 65, "prompt_tokens": 385, "total_tokens": 450, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 46, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.0142, "usd": 0.002098649350649351}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "75284cca47604e55243fd0470081fd66ec6b136ae03623b06e1e2f49473f5432"}, "error": null}
|
||||
{"id": "test-sw-000", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Baada ya kukataa kwanza ombi la Hazmi ya mkopo, msimamizi alikubali kumruhusu kutumia akaunti ya benki ya msimamizi ili kupokea uhamisho wa waya wa $5,000.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764d257fc3d6371ed318", "response_model": "kimi-k3", "response_created": 1785689678, "request_started_at": "2026-08-02T16:54:36.584963+00:00", "latency_seconds": 4.250859291991219, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Swahili: \"Baada ya kukataa kwanza ombi la Hazmi ya mkopo, msimamizi alikubali...\" Swahili is not in the listed labels, so classify as ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 78, "prompt_tokens": 409, "total_tokens": 487, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 59, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01598, "usd": 0.002361719480519481}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "4973ac8db317e5f0fda194dd28bacaf32778e9320c9d958c32976dfad57493ea"}, "error": null}
|
||||
{"id": "test-sw-001", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Kwa hiyo, watu wazima hawana haja ya kufundisha watoto wa shule ya kwanza katika kujifanya, kama wanavyofanya wakati wa kuwasaidia puzzles au kazi nyingine zinazofanana.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764d166409dfd7131599", "response_model": "kimi-k3", "response_created": 1785689678, "request_started_at": "2026-08-02T16:54:36.588018+00:00", "latency_seconds": 3.638680875010323, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Swahili: \"Kwa hiyo, watu wazima hawana haja...\" clearly Bantu/Swahili, not in listed labels. So ot. Need final exactly.", "prediction": "ot", "usage": {"completion_tokens": 62, "prompt_tokens": 405, "total_tokens": 467, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 43, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.0143, "usd": 0.0021134285714285717}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "73703f12664d18e3e4e5345d127c329f128ede1b150ca7f149dad552b50461f3"}, "error": null}
|
||||
{"id": "test-sw-002", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Ni nafasi yetu pekee... Maneno yake mengine yalipotelea kwa kelele ya mikono iliyosisitiza ya kwamba msichana yule ashikwe mateka.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764d8c789e7fd24fe8e9", "response_model": "kimi-k3", "response_created": 1785689677, "request_started_at": "2026-08-02T16:54:36.589359+00:00", "latency_seconds": 4.815949875017395, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Swahili. Swahili is not in the listed languages, so it should be classified as ot (other). Words like \"nafasi yetu\", \"maneno\", \"yalipotelea\", \"kelele\", \"msichana\", \"mateka\" are clearly Swahili/Bantu. Not in the label set. Final answer: ot.", "prediction": "ot", "usage": {"completion_tokens": 101, "prompt_tokens": 390, "total_tokens": 491, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 82, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.0179, "usd": 0.0026454805194805198}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "ef27e985be65b0ec76851222a0f1fdd1f6b87265f96badadf4ea3bffa95afc1d"}, "error": null}
|
||||
{"id": "test-sw-003", "split": "test", "source_label": "sw", "gold_label": "ot", "text": "Kwa upande mwingine, kuna majukumu kama vile mipangilio ya IT na usimamizi ambao lazima uwe ndani ya kampuni.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f764d257fc3d6371ed319", "response_model": "kimi-k3", "response_created": 1785689678, "request_started_at": "2026-08-02T16:54:36.589771+00:00", "latency_seconds": 4.621999458002392, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is Swahili: \"Kwa upande mwingine, kuna majukumu kama vile mipangilio ya IT na usimamizi ambao lazima uwe ndani ya kampuni.\" Swahili is not in the listed labels, so it should be ot. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 89, "prompt_tokens": 383, "total_tokens": 472, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 70, "rejected_prediction_tokens": null}, "prompt_tokens_details": null}, "calculated_cost": {"cny": 0.01656, "usd": 0.002447438961038961}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "df71fdbe9c50d528df4b8c6300bdee8ea923a5cc085999c38c91c956a7f5a36c"}, "error": null}
|
||||
{"id": "test-th-000", "split": "test", "source_label": "th", "gold_label": "ot", "text": "ชายหาดคือสถานที่สวยงานและเป็นสถานที่ที่ดีที่จะไปดังนั้นนั่นอาจจะเป็นหนึ่งในสถานที่โปรดปรานที่สุดของฉันที่จะไป แล้วคุณล่ะ", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7650065f01137c6c5cd9", "response_model": "kimi-k3", "response_created": 1785689681, "request_started_at": "2026-08-02T16:54:39.296325+00:00", "latency_seconds": 3.098551083006896, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "Thai, not in listed labels, so ot.", "prediction": "ot", "usage": {"completion_tokens": 31, "prompt_tokens": 429, "total_tokens": 460, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 12, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007072, "usd": 0.0010451864935064937}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "b8e79a8f9164eb6430b0497b094e2a337a806776d387ff6d87cb73f1d8caa910"}, "error": null}
|
||||
{"id": "test-th-001", "split": "test", "source_label": "th", "gold_label": "ot", "text": "เอ่อ อย่างไรก็ตามฉันสงสัยพลเมืองในมาดริดและแอตแลนตา ในขณะที่พวกเขาอาจเสียดายการสูญเสียบางส่วนของประเพณี แล้วชอบความทันสมัย", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7652e4f7240aff311c14", "response_model": "kimi-k3", "response_created": 1785689682, "request_started_at": "2026-08-02T16:54:40.166960+00:00", "latency_seconds": 5.142489125020802, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is in Thai script. Thai is not in the listed labels, so it should be classified as ot (other languages). The script is clearly Thai (เอ่อ อย่างไรก็ตามฉันสงสัย...), not any of the listed languages.", "prediction": "ot", "usage": {"completion_tokens": 82, "prompt_tokens": 438, "total_tokens": 520, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 63, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012352, "usd": 0.001825529350649351}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "eedd7d92ba6f486d05755ffe8ca201e4d23751fb0f9554b25b3653198767cbd7"}, "error": null}
|
||||
{"id": "test-th-002", "split": "test", "source_label": "th", "gold_label": "ot", "text": "บ่อยครั้งที่คนเดียวที่สามารถรักษา caida de mollera เป็น curandera", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765189b4c0360c6e96d5", "response_model": "kimi-k3", "response_created": 1785689682, "request_started_at": "2026-08-02T16:54:40.227868+00:00", "latency_seconds": 6.769287000002805, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text is mostly Thai script: \"บ่อยครั้งที่คนเดียวที่สามารถรักษา ... เป็น ...\" with Spanish words \"caida de mollera\" and \"curandera\" embedded. Deliberate mixed-language (Thai + Spanish) per rule 4 -> ot. Also Thai is outside listed set, so ot regardless. Need final exactly one line.", "prediction": "ot", "usage": {"completion_tokens": 108, "prompt_tokens": 377, "total_tokens": 485, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 89, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013732, "usd": 0.0020294825974025974}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "56ad567d45a8ecc39959245ba5909d19ccc2692a4c0717a3d3853b067f626577"}, "error": null}
|
||||
{"id": "test-th-003", "split": "test", "source_label": "th", "gold_label": "ot", "text": "Shakur ได้รับการยืนยันตัวโดยเจ้าหน้าที่ของสเปนในฐานะ Farid Hilali", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f76513cfe00996ee00051", "response_model": "kimi-k3", "response_created": 1785689682, "request_started_at": "2026-08-02T16:54:40.263350+00:00", "latency_seconds": 5.138380415999563, "attempt": 0, "content": "Final Answer: ot", "reasoning_content": "The text contains Thai script: \"ได้รับการยืนยันตัวโดยเจ้าหน้าที่ของสเปนในฐานะ\" mixed with Latin proper nouns. Thai is not in the listed labels, so classify as ot per rule 4 (language outside listed set). The visible response must be exactly Final Answer: ot.", "prediction": "ot", "usage": {"completion_tokens": 100, "prompt_tokens": 385, "total_tokens": 485, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 81, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.013092, "usd": 0.0019348955844155846}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "4067ec8965fd98cae28570b9874499b7563a2ab690f0e19da8f18112c2768b65"}, "error": null}
|
||||
{"id": "test-tr-000", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "New Republic’in Charles Lane’i, kaçırma haberinin sadece Gabriel Garcaa Marquez’in dürüst olmayan gazetecilik kayıtlarını uzattığını söylüyor.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f76527e0b44731a46e204", "response_model": "kimi-k3", "response_created": 1785689683, "request_started_at": "2026-08-02T16:54:40.686298+00:00", "latency_seconds": 5.920783665991621, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The text uses Turkish words: New Republic'in, Charles Lane'i, kaçırma haberinin, sadece, dürüst olmayan, gazetecilik kayıtlarını, uzattığını, söylüyor. Turkish suffixes and diacritics (ç, ğ, ı). Despite proper nouns and García misspelling, it's clearly Turkish. Final tr.", "prediction": "tr", "usage": {"completion_tokens": 112, "prompt_tokens": 398, "total_tokens": 510, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 93, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014552, "usd": 0.002150672207792208}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "adfc6bcff6b32fa6a66908e213bddaaec7bf189fcb9c14b4434376819da65784"}, "error": null}
|
||||
{"id": "test-tr-001", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "New Republic’in Charles Lane’i, kaçırma haberinin sadece Gabriel Garcaa Marquez’in dürüst olmayan gazetecilik kayıtlarını uzattığını söylüyor.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f76525eaf930e38c59588", "response_model": "kimi-k3", "response_created": 1785689682, "request_started_at": "2026-08-02T16:54:40.836545+00:00", "latency_seconds": 5.938103333988693, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The text: \"New Republic’in Charles Lane’i, kaçırma haberinin sadece Gabriel Garcaa Marquez’in dürüst olmayan gazetecilik kayıtlarını uzattığını söylüyor.\"\nThis is Turkish: suffixes ’in, ’i, words kaçırma, haberinin, sadece, dürüst olmayan, gazetecilik, kayıtlarını, uzattığını, söylüyor. Label tr. Need output exact line only.", "prediction": "tr", "usage": {"completion_tokens": 145, "prompt_tokens": 398, "total_tokens": 543, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 126, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.017852, "usd": 0.002638386493506494}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "adfc6bcff6b32fa6a66908e213bddaaec7bf189fcb9c14b4434376819da65784"}, "error": null}
|
||||
{"id": "test-tr-002", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "Yüzbaşı, dedi ve konuşurken, takip eden gemilere işaret etti, Albay Bishop bizi tutuyor.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765225f3623a79a1d51a", "response_model": "kimi-k3", "response_created": 1785689683, "request_started_at": "2026-08-02T16:54:41.212546+00:00", "latency_seconds": 5.128807583998423, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "The text is Turkish: \"Yüzbaşı, dedi ve konuşurken, takip eden gemilere işaret etti, Albay Bishop bizi tutuyor.\" Contains Turkish diacritics Yüzbaşı, işaret, and function words ve, eden, bizi. Proper noun Bishop doesn't change. Label tr.", "prediction": "tr", "usage": {"completion_tokens": 97, "prompt_tokens": 380, "total_tokens": 477, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 78, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012692, "usd": 0.0018757787012987014}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "64e32d475d7f617f1cb290bcd11bf5ea7e133f7452b905480edbd2576b7d16c8"}, "error": null}
|
||||
{"id": "test-tr-003", "split": "test", "source_label": "tr", "gold_label": "tr", "text": "Savaştan sonra yapılan bir örnek evlerdeki geliştirmelerde en çok bulunan şeylerden biri de çocuklardı ve çocukların şehri olarak düşünüldüğünde çok iyi tasarlanmışlardı.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7653a98ca00faff4bdcc", "response_model": "kimi-k3", "response_created": 1785689684, "request_started_at": "2026-08-02T16:54:41.405752+00:00", "latency_seconds": 4.058070166996913, "attempt": 0, "content": "Final Answer: tr", "reasoning_content": "Turkish.", "prediction": "tr", "usage": {"completion_tokens": 24, "prompt_tokens": 414, "total_tokens": 438, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 5, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006072, "usd": 0.0008973942857142858}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "3803d75a03c14ad1aaa6c54efbdfd68d7043a26807f14c290c6fae90f8cfcb9f"}, "error": null}
|
||||
{"id": "test-ur-000", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "تو ویسے ہی، والد چلا جاتا ہے اور میرے لئے چاکلیٹ دودھ کا یہ اچھا بڑا گلاس بنا دیتا ہے.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7653b59b297c3af04616", "response_model": "kimi-k3", "response_created": 1785689683, "request_started_at": "2026-08-02T16:54:42.395292+00:00", "latency_seconds": 7.514233417023206, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text is in Arabic script. Need distinguish Urdu vs Arabic. Words: تو، ویسے، والد، جاتا ہے، میرے لئے، چاکلیٹ، دودھ، گلاس. Contains Urdu-specific vocabulary and grammar: ہے، میرے لئے، جاتا ہے, چاکلیٹ دودھ. No Urdu-specific letters like ے? It has لئے with ے? Actually لئے uses ی? Urdu. Clearly Urdu. Need final exact line.", "prediction": "ur", "usage": {"completion_tokens": 129, "prompt_tokens": 386, "total_tokens": 515, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 110, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.016012, "usd": 0.0023664488311688314}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "aab6a44724fa5974b7a599f16c8fe65a313c648c34ad834a95738ccf0699ff7d"}, "error": null}
|
||||
{"id": "test-ur-001", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "مشورے اور عدالت یا انتظامی ایجنسی کی مخالفت کرنے کے لئے نوٹس بھیجا جائیں گے.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7655ec03bada3b1e8b5d", "response_model": "kimi-k3", "response_created": 1785689688, "request_started_at": "2026-08-02T16:54:45.310251+00:00", "latency_seconds": 5.579358832997968, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text is in Arabic script. Need distinguish Arabic vs Urdu. Words: مشورے (Urdu plural with ے), کے لئے (Urdu), بھیجا جائیں گے (Urdu future passive), عدالت، انتظامی ایجنسی. Presence of ے and Urdu grammar indicates Urdu. Final label ur.", "prediction": "ur", "usage": {"completion_tokens": 94, "prompt_tokens": 378, "total_tokens": 472, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 75, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.012352, "usd": 0.001825529350649351}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f4b300f21fd760aef1eaea46c2dfaaefed3ce603cd510987dfa206ea87709b1c"}, "error": null}
|
||||
{"id": "test-ur-002", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "انٹیلی جنس طور پر، یہ ممکن نہیں ہوتا کہ غیر مساوی پیچیدہ اداروں کے اس سیارے بڑے بندوق سے بڑے پیمانے پر پیدا ہوسکتے ہیں.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7656bcb8b1ca0dccff57", "response_model": "kimi-k3", "response_created": 1785689687, "request_started_at": "2026-08-02T16:54:45.402136+00:00", "latency_seconds": 4.066885042004287, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text uses Arabic script with Urdu-specific letters like ٹ (ٹ in انٹیلی), ے, ہ, and words like یہ, نہیں, ہوتا, کے, ہیں. This is Urdu.", "prediction": "ur", "usage": {"completion_tokens": 63, "prompt_tokens": 398, "total_tokens": 461, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 44, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009652, "usd": 0.0014264903896103898}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "58c83c8fcf919d8c3ad8a40e0e27b029ec4edc41d8efceaa8ac24738e178741c"}, "error": null}
|
||||
{"id": "test-ur-003", "split": "test", "source_label": "ur", "gold_label": "ur", "text": "دیکھو تم بڑی مصیبت میں آگۓ ہو", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7656a95a610865fb4d85", "response_model": "kimi-k3", "response_created": 1785689687, "request_started_at": "2026-08-02T16:54:45.464467+00:00", "latency_seconds": 7.169208542007254, "attempt": 0, "content": "Final Answer: ur", "reasoning_content": "The text is in Urdu script. It contains Urdu-specific words: دیکھو (dekho - see/look), تم (tum - you), بڑی (bari - big), مصیبت (museebat - trouble), میں (mein - in), آگۓ ہو (aagaye ho - have come/ended up). The word آگۓ with ۓ (Urdu-specific) and the overall vocabulary is distinctly Urdu, not Arabic. The phrase means \"Look, you have gotten into big trouble.\" This is clearly Urdu. Final Answer: ur.", "prediction": "ur", "usage": {"completion_tokens": 146, "prompt_tokens": 361, "total_tokens": 507, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 127, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.017212, "usd": 0.002543799480519481}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e116c203cb4c14cba886293f79a01a7a49741b05e95889b3d77f8dc57d0aaebf"}, "error": null}
|
||||
{"id": "test-vi-000", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Người câu cá nước ngọt phải có giấy phép, hãy hỏi văn phòng du lịch gần nhất để biết thông tin về cách lấy giấy phép.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765712e92b5502455056", "response_model": "kimi-k3", "response_created": 1785689688, "request_started_at": "2026-08-02T16:54:46.342059+00:00", "latency_seconds": 5.356832082994515, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text is Vietnamese: \"Người câu cá nước ngọt phải có giấy phép, hãy hỏi văn phòng du lịch gần nhất để biết thông tin về cách lấy giấy phép.\" Clear Vietnamese diacritics and vocabulary. Final answer vi.", "prediction": "vi", "usage": {"completion_tokens": 105, "prompt_tokens": 410, "total_tokens": 515, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 86, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.014092, "usd": 0.0020826877922077926}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "f77cd162909cf4663b544b92790e3fffb9c69203e55c60002157c3e554ebb290"}, "error": null}
|
||||
{"id": "test-vi-001", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Đây là kết quả mới nhất: 5,615 nhóm đối tượng tham gia hội cựu sinh viên không đóng góp, 81 1.4 phần trăm, người đóng góp nhiều nhất là $2,840 và người đóng góp ít nhất là $5.", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765725492ba8860b292b", "response_model": "kimi-k3", "response_created": 1785689688, "request_started_at": "2026-08-02T16:54:46.607481+00:00", "latency_seconds": 3.5697657089913264, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text is Vietnamese with diacritics and words like \"Đây là\", \"kết quả\", \"nhóm đối tượng\". Classify vi.", "prediction": "vi", "usage": {"completion_tokens": 58, "prompt_tokens": 442, "total_tokens": 500, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 39, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.010032, "usd": 0.0014826514285714285}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "e65e659202a76c4155df2db2f1c91e569de49f593e57b77a37e54e8296fc2ce7"}, "error": null}
|
||||
{"id": "test-vi-002", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Nghiệp vụ không chuyển đổi-- được và mất", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f7657fd361b86e42c9879", "response_model": "kimi-k3", "response_created": 1785689689, "request_started_at": "2026-08-02T16:54:46.775108+00:00", "latency_seconds": 4.782349207991501, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text is Vietnamese with diacritics: \"Nghiệp vụ không chuyển đổi-- được và mất\". Words like Nghiệp vụ, không, chuyển đổi, được, và, mất are Vietnamese. Final answer vi.", "prediction": "vi", "usage": {"completion_tokens": 85, "prompt_tokens": 365, "total_tokens": 450, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 66, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.011192, "usd": 0.00165409038961039}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "5a8f276c0f4f9025eeebebb4b54a194c3dae71afa119c7ff421769b4d6b81f93"}, "error": null}
|
||||
{"id": "test-vi-003", "split": "test", "source_label": "vi", "gold_label": "vi", "text": "Câu hỏi duy nhất về cuộc khảo sát NEA là Bạn đã đọc bất kỳ tài liệu nào trong năm qua chưa?", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765789b4c0360c6e96e7", "response_model": "kimi-k3", "response_created": 1785689688, "request_started_at": "2026-08-02T16:54:46.997624+00:00", "latency_seconds": 6.4159510000026785, "attempt": 0, "content": "Final Answer: vi", "reasoning_content": "The text is Vietnamese: \"Câu hỏi duy nhất về cuộc khảo sát NEA là Bạn đã đọc bất kỳ tài liệu nào trong năm qua chưa?\" Contains Vietnamese diacritics and words like Câu hỏi, duy nhất, về, cuộc, khảo sát, Bạn đã đọc, bất kỳ, tài liệu, nào, trong năm qua chưa. NEA is an acronym/proper noun ignore. Label vi. Need final exact line only.", "prediction": "vi", "usage": {"completion_tokens": 164, "prompt_tokens": 396, "total_tokens": 560, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 145, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.019712, "usd": 0.0029132800000000003}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "81a4f6c6df78b9ef59ff5c85b4cb3ed3be9cece741cab816b5f28f67d65829ac"}, "error": null}
|
||||
{"id": "test-zh-000", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "根本不是那么回事，用指甲都能划出痕迹 用钥匙那叫一个惨不忍睹啊", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765aa98ca00faff4bde0", "response_model": "kimi-k3", "response_created": 1785689691, "request_started_at": "2026-08-02T16:54:49.469631+00:00", "latency_seconds": 3.3357121250010096, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is Chinese (Simplified): 根本不是那么回事，用指甲都能划出痕迹 用钥匙那叫一个惨不忍睹啊. Classify zh. Need final exact one line.", "prediction": "zh", "usage": {"completion_tokens": 55, "prompt_tokens": 360, "total_tokens": 415, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 36, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.008092, "usd": 0.0011959345454545455}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "211ebe777925d94f8f011b3a4e6dfbf61a2734632ecdb79ec7c953e3c7d81d67"}, "error": null}
|
||||
{"id": "test-zh-001", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "邮寄地址错误，请尽快处理，上次回复说是调查清楚后给我回电，又过去一个礼拜，一点消息没有！", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765bfcd3147de4645819", "response_model": "kimi-k3", "response_created": 1785689692, "request_started_at": "2026-08-02T16:54:49.910057+00:00", "latency_seconds": 4.106191166996723, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is Chinese (Simplified): \"邮寄地址错误，请尽快处理...\" Clear zh. Need output exactly one line.", "prediction": "zh", "usage": {"completion_tokens": 46, "prompt_tokens": 371, "total_tokens": 417, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 27, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.007412, "usd": 0.0010954358441558443}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "d9cae906bbb10a4b1b25dd21c4bdfcae23b22138709cfd0cae7b977c6c831b17"}, "error": null}
|
||||
{"id": "test-zh-002", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "面料很好，透气舒服，但是档部设计不好，穿着很难受，不像是可以直接穿的内裤", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765ae4f7240aff311c26", "response_model": "kimi-k3", "response_created": 1785689691, "request_started_at": "2026-08-02T16:54:50.177569+00:00", "latency_seconds": 2.6525153330003377, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The text is Chinese characters, Simplified. Classify zh. Final exactly one line.", "prediction": "zh", "usage": {"completion_tokens": 38, "prompt_tokens": 365, "total_tokens": 403, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 19, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.006492, "usd": 0.0009594670129870131}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "8e2a1bb42ccdf0b0657627fd271e7144871148a6c39cc7cb66b7201cbfdc0f72"}, "error": null}
|
||||
{"id": "test-zh-003", "split": "test", "source_label": "zh", "gold_label": "zh", "text": "我订购的是独家签名版，为何收到的书是无签名的？？", "provider": "moonshot", "base_url": "https://api.moonshot.cn/v1", "model_requested": "kimi-k3", "response_id": "chatcmpl-6a6f765b3f3c9b5f060c0d69", "response_model": "kimi-k3", "response_created": 1785689693, "request_started_at": "2026-08-02T16:54:50.889960+00:00", "latency_seconds": 5.634893791022478, "attempt": 0, "content": "Final Answer: zh", "reasoning_content": "The user wants classification. Need output exactly final answer. Text is Chinese Simplified: 我订购的是独家签名版，为何收到的书是无签名的？？ Means I ordered exclusive signed edition, why received book unsigned. Label zh. Need final exactly one line.", "prediction": "zh", "usage": {"completion_tokens": 73, "prompt_tokens": 359, "total_tokens": 432, "completion_tokens_details": {"accepted_prediction_tokens": null, "audio_tokens": null, "reasoning_tokens": 54, "rejected_prediction_tokens": null}, "prompt_tokens_details": {"audio_tokens": null, "cache_write_tokens": null, "cached_tokens": 256}, "cached_tokens": 256}, "calculated_cost": {"cny": 0.009872, "usd": 0.0014590046753246756}, "request": {"max_tokens": 256, "reasoning_effort": "low", "prompt_sha256": "afd86326c0dc51de1119f55e32ea9844b43f1cf6b6d0a983927a18c6451d07e0"}, "error": null}
|
||||
+50
@@ -0,0 +1,50 @@
|
||||
{
|
||||
"teacher": {
|
||||
"provider": "moonshot",
|
||||
"model": "kimi-k3",
|
||||
"pricing": {
|
||||
"input_per_million": 20.0,
|
||||
"cached_input_per_million": 2.0,
|
||||
"output_per_million": 100.0,
|
||||
"currency": "CNY",
|
||||
"source_url": "https://platform.kimi.com/docs/pricing/chat-k3.md",
|
||||
"as_of": "2026-07-29",
|
||||
"usd_per_currency_unit": 0.1477922077922078,
|
||||
"fx_source_url": "https://www.ecb.europa.eu/stats/eurofxref/eurofxref-daily.xml",
|
||||
"fx_as_of": "2026-07-29"
|
||||
}
|
||||
},
|
||||
"train": {
|
||||
"rows": 160,
|
||||
"retained_receipts": 160,
|
||||
"coverage": 1.0,
|
||||
"valid_receipts": 160,
|
||||
"unique_response_ids": 160,
|
||||
"correct": 160,
|
||||
"gold_accuracy": 1.0,
|
||||
"observed_gold_accuracy": 1.0,
|
||||
"prompt_tokens": 63569,
|
||||
"completion_tokens": 12503,
|
||||
"latency_seconds_total": 629.3293080376461,
|
||||
"latency_seconds_mean": 3.9333081752352883,
|
||||
"provider_cost_cny": 1.867344,
|
||||
"provider_cost_usd": 0.27597889246753254
|
||||
},
|
||||
"test": {
|
||||
"rows": 80,
|
||||
"retained_receipts": 80,
|
||||
"coverage": 1.0,
|
||||
"valid_receipts": 80,
|
||||
"unique_response_ids": 80,
|
||||
"correct": 80,
|
||||
"gold_accuracy": 1.0,
|
||||
"observed_gold_accuracy": 1.0,
|
||||
"prompt_tokens": 30898,
|
||||
"completion_tokens": 6325,
|
||||
"latency_seconds_total": 331.5117934176815,
|
||||
"latency_seconds_mean": 4.143897417721019,
|
||||
"provider_cost_cny": 0.9186840000000001,
|
||||
"provider_cost_usd": 0.13577433662337665
|
||||
},
|
||||
"campaign_complete": true
|
||||
}
|
||||
+230
@@ -0,0 +1,230 @@
|
||||
{
|
||||
"started_at": "2026-08-01T06:41:47.450705+00:00",
|
||||
"finished_at": "2026-08-01T06:41:54.747196+00:00",
|
||||
"runtime_seconds": 7.059750179760158,
|
||||
"device": "cuda",
|
||||
"torch_version": "2.11.0+cu130",
|
||||
"base_model": "HuggingFaceTB/SmolLM2-135M-Instruct",
|
||||
"base_model_revision": "12fd25f77366fa6b3b4b768ec3050bf629380bac",
|
||||
"training_rows": 160,
|
||||
"epochs": 5,
|
||||
"batch_size": 4,
|
||||
"optimizer": "AdamW",
|
||||
"learning_rate": 0.0008,
|
||||
"lora": {
|
||||
"r": 16,
|
||||
"alpha": 32,
|
||||
"targets": [
|
||||
"q_proj",
|
||||
"v_proj"
|
||||
]
|
||||
},
|
||||
"trainable_parameters": 921600,
|
||||
"total_parameters_with_adapter": 135436608,
|
||||
"optimizer_steps": 200,
|
||||
"first_loss": 12.23326587677002,
|
||||
"final_loss": 0.0023392525035887957,
|
||||
"mean_loss": 0.6463202222852124,
|
||||
"losses": [
|
||||
12.23326587677002,
|
||||
10.062297821044922,
|
||||
10.695526123046875,
|
||||
7.681145191192627,
|
||||
5.883533000946045,
|
||||
5.580435276031494,
|
||||
4.073431015014648,
|
||||
3.567901372909546,
|
||||
3.2300102710723877,
|
||||
2.6896257400512695,
|
||||
2.388627052307129,
|
||||
1.8905359506607056,
|
||||
2.320594310760498,
|
||||
2.5594756603240967,
|
||||
2.0164566040039062,
|
||||
2.567136526107788,
|
||||
2.5213701725006104,
|
||||
2.2410974502563477,
|
||||
1.7878096103668213,
|
||||
1.2679821252822876,
|
||||
1.6270002126693726,
|
||||
1.1315091848373413,
|
||||
0.9897012114524841,
|
||||
2.3076086044311523,
|
||||
1.4622139930725098,
|
||||
1.6360044479370117,
|
||||
0.673614501953125,
|
||||
1.1829167604446411,
|
||||
1.5917915105819702,
|
||||
1.198855996131897,
|
||||
1.4503737688064575,
|
||||
0.428786039352417,
|
||||
0.20316383242607117,
|
||||
0.24510720372200012,
|
||||
0.5218604207038879,
|
||||
1.1607334613800049,
|
||||
0.5434024930000305,
|
||||
0.7045337557792664,
|
||||
1.0069295167922974,
|
||||
0.47515979409217834,
|
||||
0.9643514752388,
|
||||
0.1116938665509224,
|
||||
0.2660924196243286,
|
||||
0.3425658941268921,
|
||||
0.8563904762268066,
|
||||
0.7196137309074402,
|
||||
1.0455775260925293,
|
||||
0.9870248436927795,
|
||||
0.1886233240365982,
|
||||
0.7767462134361267,
|
||||
0.6876943111419678,
|
||||
0.6249509453773499,
|
||||
0.5260353088378906,
|
||||
0.42842400074005127,
|
||||
0.20938293635845184,
|
||||
0.2536408305168152,
|
||||
0.683778703212738,
|
||||
0.6062668561935425,
|
||||
0.21564440429210663,
|
||||
0.6588385105133057,
|
||||
0.7190943360328674,
|
||||
0.09460505098104477,
|
||||
0.5058103799819946,
|
||||
0.9891417026519775,
|
||||
0.38562265038490295,
|
||||
0.5294267535209656,
|
||||
0.02541974186897278,
|
||||
0.4129575788974762,
|
||||
0.7445634007453918,
|
||||
0.30863261222839355,
|
||||
0.022321775555610657,
|
||||
0.18306231498718262,
|
||||
0.14176572859287262,
|
||||
0.21323193609714508,
|
||||
0.023611346259713173,
|
||||
0.3535572290420532,
|
||||
0.10225439071655273,
|
||||
0.5266523957252502,
|
||||
0.07125026732683182,
|
||||
0.029830845072865486,
|
||||
0.2521970272064209,
|
||||
0.3975638151168823,
|
||||
0.04321639984846115,
|
||||
0.07781067490577698,
|
||||
0.020981086418032646,
|
||||
0.021831966936588287,
|
||||
0.008844907395541668,
|
||||
0.004404002334922552,
|
||||
0.0020085813011974096,
|
||||
0.010316853411495686,
|
||||
0.11901021003723145,
|
||||
0.14854948222637177,
|
||||
0.008009753189980984,
|
||||
0.06145263835787773,
|
||||
0.0010108593851327896,
|
||||
0.01561696082353592,
|
||||
0.08826116472482681,
|
||||
0.029440516605973244,
|
||||
0.052831560373306274,
|
||||
0.014964704401791096,
|
||||
0.17732924222946167,
|
||||
0.008418903686106205,
|
||||
0.3147701323032379,
|
||||
0.2750137448310852,
|
||||
0.09627381712198257,
|
||||
0.004877119790762663,
|
||||
0.024520136415958405,
|
||||
0.04009673371911049,
|
||||
0.023672880604863167,
|
||||
0.004260607063770294,
|
||||
0.007769071031361818,
|
||||
0.00630973419174552,
|
||||
0.006435201037675142,
|
||||
0.008410915732383728,
|
||||
0.298997163772583,
|
||||
0.0033175835851579905,
|
||||
0.011488684453070164,
|
||||
0.09618687629699707,
|
||||
0.08273980021476746,
|
||||
0.11833988130092621,
|
||||
0.04111325368285179,
|
||||
0.0009924803161993623,
|
||||
0.06604669243097305,
|
||||
0.010862606577575207,
|
||||
0.18383924663066864,
|
||||
0.002522931434214115,
|
||||
0.004025444854050875,
|
||||
0.18946759402751923,
|
||||
0.0072350818663835526,
|
||||
0.0076723527163267136,
|
||||
0.0041358694434165955,
|
||||
0.010324793867766857,
|
||||
0.0018014800734817982,
|
||||
0.003930415026843548,
|
||||
0.006519278045743704,
|
||||
0.003409093478694558,
|
||||
0.10473594069480896,
|
||||
0.0019360095029696822,
|
||||
0.004570731893181801,
|
||||
0.007261015474796295,
|
||||
0.0017995196394622326,
|
||||
0.0006797119858674705,
|
||||
0.0024804044514894485,
|
||||
0.019872531294822693,
|
||||
0.00176086591091007,
|
||||
0.001979110762476921,
|
||||
0.0043699792586266994,
|
||||
0.0011555724777281284,
|
||||
0.0023414648603647947,
|
||||
0.0007984342519193888,
|
||||
0.0018523266771808267,
|
||||
0.001224210485816002,
|
||||
0.008516449481248856,
|
||||
0.0011650845408439636,
|
||||
0.0013863900676369667,
|
||||
0.001624621800146997,
|
||||
0.02259969525039196,
|
||||
0.002657415112480521,
|
||||
0.0015202113427221775,
|
||||
0.0508158802986145,
|
||||
0.002426866674795747,
|
||||
0.004635736346244812,
|
||||
0.000522429239936173,
|
||||
0.0005015835049562156,
|
||||
0.004379968158900738,
|
||||
0.0018349788151681423,
|
||||
0.0010437751188874245,
|
||||
0.0013649024767801166,
|
||||
0.004170037806034088,
|
||||
0.01009396743029356,
|
||||
0.004265328403562307,
|
||||
0.011878587305545807,
|
||||
0.0013235677033662796,
|
||||
0.0018509732326492667,
|
||||
0.001259719836525619,
|
||||
0.0010441394988447428,
|
||||
0.0008565496536903083,
|
||||
0.002013395307585597,
|
||||
0.0003855683025904,
|
||||
0.00021147385996300727,
|
||||
0.0005270058754831553,
|
||||
0.0014656938146799803,
|
||||
0.0003616194298956543,
|
||||
0.0020853427704423666,
|
||||
0.0002984879829455167,
|
||||
0.005857599899172783,
|
||||
0.060871634632349014,
|
||||
0.0005054053035564721,
|
||||
0.0016232342459261417,
|
||||
0.00027615882572717965,
|
||||
0.0012551458785310388,
|
||||
0.0008329860283993185,
|
||||
0.0003348049649503082,
|
||||
0.00284484401345253,
|
||||
0.0009668312850408256,
|
||||
0.0040282015688717365,
|
||||
0.0013629095628857613,
|
||||
0.0004559664230328053,
|
||||
0.003461322980001569,
|
||||
0.0023392525035887957
|
||||
]
|
||||
}
|
||||
Reference in New Issue
Block a user