liqiang
b119135836
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
22 lines
1.8 KiB
Markdown
22 lines
1.8 KiB
Markdown
# 实验 8-19：特殊字符串的精确复述 SFT
|
||
|
||
本实验从 Coding Agent 的 `old_string` 匹配失败出发，区分模型复制错误与 tokenizer、序列化、Harness 和工具层错误，然后用未见过的随机字符串和工具参数任务训练 Qwen3-8B 的 LoRA 适配器。训练目标是 byte-exact 复制，而不是语义相似或“看起来一样”。
|
||
|
||
## 运行
|
||
|
||
```bash
|
||
cd chapter8/exact-copy-sft
|
||
python generate_data.py
|
||
python train_sft.py --model Qwen/Qwen3-8B
|
||
python evaluate.py --model Qwen/Qwen3-8B --adapter output/adapter
|
||
python tokenizer_audit.py
|
||
```
|
||
|
||
训练和评估必须使用本机 CUDA GPU 与开源 Hugging Face 模型。数据按随机种子、字符串长度、token 组合和上下文包装隔离；`validation/` 保存训练回执和独立回归报告。
|
||
|
||
扩容后先对三类任务、10 种语言上下文和 8 种文章体裁做分层人工抽查，记录见 [`validation/manual_audit.md`](validation/manual_audit.md)；再运行 tokenizer 审计，避免把 tokenizer/序列化损坏误判为模型能力问题。
|
||
|
||
如果模型在直接复述探针中正确、但工具调用仍失败，应修复 Harness 或工具协议，不应把系统层损坏误报为后训练收益。
|
||
|
||
本机 RTX PRO 6000 实测：1024 条训练样本、256 条留出样本、256 条边界样本，训练 2 个 epoch、Qwen3-8B bf16 LoRA。留出集 byte-exact accuracy 从基座 37.5% 提升到 78.9%，独立边界集为 80.1%；平均首次字节分歧位置分别为 54.0 和 54.2。另用 512 条留出/边界探针审计 3 个开源 tokenizer：Qwen3 与 Qwen2.5 round-trip 均为 80.1%，Mistral 为 100%，说明 tokenizer 层也必须单独设回归门禁。结果文件见 `validation/eval_base_eval.json`、`validation/eval_adapted_eval.json`、`validation/eval_adapted_boundary.json` 和 `validation/tokenizer_audit.json`。
|