liqiang
b119135836
33 KiB
Cursor Chat: ai-agent-book
Metadata
- Project: ai-agent-book
- Path:
/Users/boj - Date: 2025-10-01 12:31:22
- Session ID:
2a1556e8-109d-4f11-8f86-614c97698c26
Conversation
👤 You
(TaskRunner pid=153792) 'profiler': {'target': 'verl.utils.profiler.ProfilerConfig',
(TaskRunner pid=153792) 'all_ranks': False,
(TaskRunner pid=153792) 'enable': False,
(TaskRunner pid=153792) 'ranks': [],
(TaskRunner pid=153792) 'save_path': 'outputs/profile',
(TaskRunner pid=153792) 'tool': None,
(TaskRunner pid=153792) 'tool_config': {'npu': {'target': 'verl.utils.profiler.config.NPUToolConfig',
(TaskRunner pid=153792) 'analysis': True,
(TaskRunner pid=153792) 'contents': [],
(TaskRunner pid=153792) 'discrete': False,
(TaskRunner pid=153792) 'level': 'level1'},
(TaskRunner pid=153792) 'nsys': {'target': 'verl.utils.profiler.config.NsightToolConfig',
(TaskRunner pid=153792) 'discrete': False},
(TaskRunner pid=153792) 'torch': {'target': 'verl.utils.profiler.config.TorchProfilerToolConfig',
(TaskRunner pid=153792) 'step_end': None,
(TaskRunner pid=153792) 'step_start': 0},
(TaskRunner pid=153792) 'torch_memory': {'target': 'verl.utils.profiler.config.TorchMemoryToolConfig',
(TaskRunner pid=153792) 'stack_depth': 32,
(TaskRunner pid=153792) 'trace_alloc_max_entries': 100000}}},
(TaskRunner pid=153792) 'prompt_length': 2048,
(TaskRunner pid=153792) 'response_length': 16384,
(TaskRunner pid=153792) 'skip_dump_dir': '/tmp/rollout_dump',
(TaskRunner pid=153792) 'skip_rollout': False,
(TaskRunner pid=153792) 'skip_tokenizer_init': True,
(TaskRunner pid=153792) 'temperature': 1.0,
(TaskRunner pid=153792) 'tensor_model_parallel_size': 4,
(TaskRunner pid=153792) 'top_k': -1,
(TaskRunner pid=153792) 'top_p': 1,
(TaskRunner pid=153792) 'trace': {'target': 'verl.workers.config.TraceConfig',
(TaskRunner pid=153792) 'backend': None,
(TaskRunner pid=153792) 'token2text': False},
(TaskRunner pid=153792) 'update_weights_bucket_megabytes': 512,
(TaskRunner pid=153792) 'val_kwargs': {'target': 'verl.workers.config.SamplingConfig',
(TaskRunner pid=153792) 'do_sample': False,
(TaskRunner pid=153792) 'n': 30,
(TaskRunner pid=153792) 'temperature': 1.0,
(TaskRunner pid=153792) 'top_k': -1,
(TaskRunner pid=153792) 'top_p': 0.6}}},
(TaskRunner pid=153792) 'algorithm': {'target': 'verl.trainer.config.AlgoConfig',
(TaskRunner pid=153792) 'adv_estimator': 'grpo',
(TaskRunner pid=153792) 'gamma': 1.0,
(TaskRunner pid=153792) 'kl_ctrl': {'target': 'verl.trainer.config.KLControlConfig',
(TaskRunner pid=153792) 'horizon': 10000,
(TaskRunner pid=153792) 'kl_coef': 0.0,
(TaskRunner pid=153792) 'target_kl': 0.1,
(TaskRunner pid=153792) 'type': 'fixed'},
(TaskRunner pid=153792) 'kl_penalty': 'kl',
(TaskRunner pid=153792) 'lam': 1.0,
(TaskRunner pid=153792) 'norm_adv_by_std_in_grpo': True,
(TaskRunner pid=153792) 'pf_ppo': {'reweight_method': 'pow', 'weight_pow': 2.0},
(TaskRunner pid=153792) 'use_kl_in_reward': False,
(TaskRunner pid=153792) 'use_pf_ppo': False},
(TaskRunner pid=153792) 'critic': {'target': 'verl.workers.config.FSDPCriticConfig',
(TaskRunner pid=153792) 'checkpoint': {'target': 'verl.trainer.config.CheckpointConfig',
(TaskRunner pid=153792) 'async_save': False,
(TaskRunner pid=153792) 'load_contents': ['model', 'optimizer', 'extra'],
(TaskRunner pid=153792) 'save_contents': ['model', 'optimizer', 'extra']},
(TaskRunner pid=153792) 'cliprange_value': 0.5,
(TaskRunner pid=153792) 'enable': None,
(TaskRunner pid=153792) 'forward_max_token_len_per_gpu': 32768,
(TaskRunner pid=153792) 'forward_micro_batch_size': None,
(TaskRunner pid=153792) 'forward_micro_batch_size_per_gpu': None,
(TaskRunner pid=153792) 'grad_clip': 1.0,
(TaskRunner pid=153792) 'loss_agg_mode': 'token-mean',
(TaskRunner pid=153792) 'model': {'target': 'verl.workers.config.FSDPCriticModelCfg',
(TaskRunner pid=153792) 'enable_activation_offload': False,
(TaskRunner pid=153792) 'enable_gradient_checkpointing': True,
(TaskRunner pid=153792) 'external_lib': None,
(TaskRunner pid=153792) 'fsdp_config': {'target': 'verl.workers.config.FSDPEngineConfig',
(TaskRunner pid=153792) 'entropy_checkpointing': False,
(TaskRunner pid=153792) 'entropy_from_logits_with_chunking': False,
(TaskRunner pid=153792) 'forward_only': False,
(TaskRunner pid=153792) 'forward_prefetch': False,
(TaskRunner pid=153792) 'fsdp_size': -1,
(TaskRunner pid=153792) 'model_dtype': 'fp32',
(TaskRunner pid=153792) 'offload_policy': False,
(TaskRunner pid=153792) 'optimizer_offload': False,
(TaskRunner pid=153792) 'param_offload': False,
(TaskRunner pid=153792) 'reshard_after_forward': True,
(TaskRunner pid=153792) 'strategy': 'fsdp',
(TaskRunner pid=153792) 'ulysses_sequence_parallel_size': 1,
(TaskRunner pid=153792) 'use_orig_params': False,
(TaskRunner pid=153792) 'use_torch_compile': True,
(TaskRunner pid=153792) 'wrap_policy': {'min_num_params': 0}},
(TaskRunner pid=153792) 'lora_alpha': 16,
(TaskRunner pid=153792) 'lora_rank': 0,
(TaskRunner pid=153792) 'override_config': {},
(TaskRunner pid=153792) 'path': '/models/deepseek-llm-7b-chat',
(TaskRunner pid=153792) 'target_modules': 'all-linear',
(TaskRunner pid=153792) 'tokenizer_path': '/root/verl/recipe/retool/checkpoint/multiturn-sft-qwen-2.5-32b-instruct/global_step_372',
(TaskRunner pid=153792) 'trust_remote_code': False,
(TaskRunner pid=153792) 'use_remove_padding': False,
(TaskRunner pid=153792) 'use_shm': False},
(TaskRunner pid=153792) 'optim': {'target': 'verl.workers.config.FSDPOptimizerConfig',
(TaskRunner pid=153792) 'betas': [0.9, 0.999],
(TaskRunner pid=153792) 'clip_grad': 1.0,
(TaskRunner pid=153792) 'lr': 1e-05,
(TaskRunner pid=153792) 'lr_warmup_steps': -1,
(TaskRunner pid=153792) 'lr_warmup_steps_ratio': 0.0,
(TaskRunner pid=153792) 'min_lr_ratio': 0.0,
(TaskRunner pid=153792) 'num_cycles': 0.5,
(TaskRunner pid=153792) 'total_training_steps': -1,
(TaskRunner pid=153792) 'warmup_style': 'constant',
(TaskRunner pid=153792) 'weight_decay': 0.01},
(TaskRunner pid=153792) 'ppo_epochs': 1,
(TaskRunner pid=153792) 'ppo_max_token_len_per_gpu': 32768,
(TaskRunner pid=153792) 'ppo_micro_batch_size': None,
(TaskRunner pid=153792) 'ppo_micro_batch_size_per_gpu': None,
(TaskRunner pid=153792) 'ppo_mini_batch_size': 64,
(TaskRunner pid=153792) 'profiler': {'target': 'verl.utils.profiler.ProfilerConfig',
(TaskRunner pid=153792) 'all_ranks': False,
(TaskRunner pid=153792) 'enable': False,
(TaskRunner pid=153792) 'ranks': [],
(TaskRunner pid=153792) 'save_path': 'outputs/profile',
(TaskRunner pid=153792) 'tool': None,
(TaskRunner pid=153792) 'tool_config': {'npu': {'target': 'verl.utils.profiler.config.NPUToolConfig',
(TaskRunner pid=153792) 'analysis': True,
(TaskRunner pid=153792) 'contents': [],
(TaskRunner pid=153792) 'discrete': False,
(TaskRunner pid=153792) 'level': 'level1'},
(TaskRunner pid=153792) 'nsys': {'target': 'verl.utils.profiler.config.NsightToolConfig',
(TaskRunner pid=153792) 'discrete': False},
(TaskRunner pid=153792) 'torch': {'target': 'verl.utils.profiler.config.TorchProfilerToolConfig',
(TaskRunner pid=153792) 'step_end': None,
(TaskRunner pid=153792) 'step_start': 0},
(TaskRunner pid=153792) 'torch_memory': {'target': 'verl.utils.profiler.config.TorchMemoryToolConfig',
(TaskRunner pid=153792) 'stack_depth': 32,
(TaskRunner pid=153792) 'trace_alloc_max_entries': 100000}}},
(TaskRunner pid=153792) 'rollout_n': 16,
(TaskRunner pid=153792) 'shuffle': False,
(TaskRunner pid=153792) 'strategy': 'fsdp',
(TaskRunner pid=153792) 'ulysses_sequence_parallel_size': 1,
(TaskRunner pid=153792) 'use_dynamic_bsz': True},
(TaskRunner pid=153792) 'custom_reward_function': {'name': 'compute_score',
(TaskRunner pid=153792) 'path': 'recipe/retool/retool.py'},
(TaskRunner pid=153792) 'data': {'apply_chat_template_kwargs': {},
(TaskRunner pid=153792) 'custom_cls': {'name': 'CustomRLHFDataset',
(TaskRunner pid=153792) 'path': 'recipe/retool/retool.py'},
(TaskRunner pid=153792) 'datagen': {'name': None, 'path': None},
(TaskRunner pid=153792) 'dataloader_num_workers': 8,
(TaskRunner pid=153792) 'filter_overlong_prompts': True,
(TaskRunner pid=153792) 'filter_overlong_prompts_workers': 1,
(TaskRunner pid=153792) 'image_key': 'images',
(TaskRunner pid=153792) 'max_prompt_length': 2048,
(TaskRunner pid=153792) 'max_response_length': 16384,
(TaskRunner pid=153792) 'prompt_key': 'prompt',
(TaskRunner pid=153792) 'return_full_prompt': False,
(TaskRunner pid=153792) 'return_multi_modal_inputs': True,
(TaskRunner pid=153792) 'return_raw_chat': True,
(TaskRunner pid=153792) 'return_raw_input_ids': False,
(TaskRunner pid=153792) 'reward_fn_key': 'data_source',
(TaskRunner pid=153792) 'sampler': {'class_name': None, 'class_path': None},
(TaskRunner pid=153792) 'shuffle': True,
(TaskRunner pid=153792) 'tokenizer': None,
(TaskRunner pid=153792) 'train_batch_size': 512,
(TaskRunner pid=153792) 'train_files': ['/dataset/BytedTsinghua-SIA/DAPO-Math-17k'],
(TaskRunner pid=153792) 'truncation': 'error',
(TaskRunner pid=153792) 'trust_remote_code': False,
(TaskRunner pid=153792) 'use_shm': False,
(TaskRunner pid=153792) 'val_batch_size': None,
(TaskRunner pid=153792) 'val_files': ['/dataset/yentinglin/aime_2025'],
(TaskRunner pid=153792) 'validation_shuffle': False,
(TaskRunner pid=153792) 'video_key': 'videos'},
(TaskRunner pid=153792) 'global_profiler': {'target': 'verl.utils.profiler.ProfilerConfig',
(TaskRunner pid=153792) 'global_tool_config': {'nsys': {'target': 'verl.utils.profiler.config.NsightToolConfig',
(TaskRunner pid=153792) 'controller_nsight_options': {'cuda-graph-trace': 'graph',
(TaskRunner pid=153792) 'cuda-memory-usage': 'true',
(TaskRunner pid=153792) 'trace': 'cuda,nvtx,cublas,ucx'},
(TaskRunner pid=153792) 'discrete': False,
(TaskRunner pid=153792) 'worker_nsight_options': {'capture-range': 'cudaProfilerApi',
(TaskRunner pid=153792) 'capture-range-end': None,
(TaskRunner pid=153792) 'cuda-graph-trace': 'graph',
(TaskRunner pid=153792) 'cuda-memory-usage': 'true',
(TaskRunner pid=153792) 'kill': 'none',
(TaskRunner pid=153792) 'trace': 'cuda,nvtx,cublas,ucx'}},
(TaskRunner pid=153792) 'torch_memory': {'context': 'all',
(TaskRunner pid=153792) 'kw_args': {},
(TaskRunner pid=153792) 'stack_depth': 32,
(TaskRunner pid=153792) 'stacks': 'all',
(TaskRunner pid=153792) 'trace_alloc_max_entries': 100000}},
(TaskRunner pid=153792) 'profile_continuous_steps': False,
(TaskRunner pid=153792) 'save_path': 'outputs/profile',
(TaskRunner pid=153792) 'steps': None,
(TaskRunner pid=153792) 'tool': None},
(TaskRunner pid=153792) 'ray_kwargs': {'ray_init': {'num_cpus': None}, 'timeline_json_file': None},
(TaskRunner pid=153792) 'reward_model': {'enable': False,
(TaskRunner pid=153792) 'enable_resource_pool': False,
(TaskRunner pid=153792) 'forward_max_token_len_per_gpu': 32768,
(TaskRunner pid=153792) 'launch_reward_fn_async': False,
(TaskRunner pid=153792) 'max_length': None,
(TaskRunner pid=153792) 'micro_batch_size': None,
(TaskRunner pid=153792) 'micro_batch_size_per_gpu': None,
(TaskRunner pid=153792) 'model': {'external_lib': None,
(TaskRunner pid=153792) 'fsdp_config': {'target': 'verl.workers.config.FSDPEngineConfig',
(TaskRunner pid=153792) 'forward_prefetch': False,
(TaskRunner pid=153792) 'fsdp_size': -1,
(TaskRunner pid=153792) 'param_offload': False,
(TaskRunner pid=153792) 'reshard_after_forward': True,
(TaskRunner pid=153792) 'wrap_policy': {'min_num_params': 0}},
(TaskRunner pid=153792) 'input_tokenizer': '/root/verl/recipe/retool/checkpoint/multiturn-sft-qwen-2.5-32b-instruct/global_step_372',
(TaskRunner pid=153792) 'path': '/models/FsfairX-LLaMA3-RM-v0.1',
(TaskRunner pid=153792) 'trust_remote_code': False,
(TaskRunner pid=153792) 'use_fused_kernels': False,
(TaskRunner pid=153792) 'use_remove_padding': False,
(TaskRunner pid=153792) 'use_shm': False},
(TaskRunner pid=153792) 'n_gpus_per_node': 0,
(TaskRunner pid=153792) 'nnodes': 0,
(TaskRunner pid=153792) 'profiler': {'target': 'verl.utils.profiler.ProfilerConfig',
(TaskRunner pid=153792) 'all_ranks': False,
(TaskRunner pid=153792) 'enable': False,
(TaskRunner pid=153792) 'ranks': [],
(TaskRunner pid=153792) 'save_path': 'outputs/profile',
(TaskRunner pid=153792) 'tool': None,
(TaskRunner pid=153792) 'tool_config': {'npu': {'target': 'verl.utils.profiler.config.NPUToolConfig',
(TaskRunner pid=153792) 'analysis': True,
(TaskRunner pid=153792) 'contents': [],
(TaskRunner pid=153792) 'discrete': False,
(TaskRunner pid=153792) 'level': 'level1'},
(TaskRunner pid=153792) 'nsys': {'target': 'verl.utils.profiler.config.NsightToolConfig',
(TaskRunner pid=153792) 'discrete': False},
(TaskRunner pid=153792) 'torch': {'target': 'verl.utils.profiler.config.TorchProfilerToolConfig',
(TaskRunner pid=153792) 'step_end': None,
(TaskRunner pid=153792) 'step_start': 0},
(TaskRunner pid=153792) 'torch_memory': {'target': 'verl.utils.profiler.config.TorchMemoryToolConfig',
(TaskRunner pid=153792) 'stack_depth': 32,
(TaskRunner pid=153792) 'trace_alloc_max_entries': 100000}}},
(TaskRunner pid=153792) 'reward_manager': 'naive',
(TaskRunner pid=153792) 'sandbox_fusion': {'max_concurrent': 64,
(TaskRunner pid=153792) 'memory_limit_mb': 1024,
(TaskRunner pid=153792) 'url': None},
(TaskRunner pid=153792) 'strategy': 'fsdp',
(TaskRunner pid=153792) 'ulysses_sequence_parallel_size': 1,
(TaskRunner pid=153792) 'use_dynamic_bsz': True},
(TaskRunner pid=153792) 'trainer': {'balance_batch': True,
(TaskRunner pid=153792) 'critic_warmup': 0,
(TaskRunner pid=153792) 'default_hdfs_dir': None,
(TaskRunner pid=153792) 'default_local_dir': '/root/verl/recipe/retool/checkpoint/qwen2.5-32b_dapo',
(TaskRunner pid=153792) 'del_local_ckpt_after_load': False,
(TaskRunner pid=153792) 'device': 'cuda',
(TaskRunner pid=153792) 'esi_redundant_time': 0,
(TaskRunner pid=153792) 'experiment_name': 'qwen2.5-32b_dapo',
(TaskRunner pid=153792) 'log_val_generations': 100,
(TaskRunner pid=153792) 'logger': ['console', 'wandb'],
(TaskRunner pid=153792) 'max_actor_ckpt_to_keep': None,
(TaskRunner pid=153792) 'max_critic_ckpt_to_keep': None,
(TaskRunner pid=153792) 'n_gpus_per_node': 8,
(TaskRunner pid=153792) 'nnodes': 1,
(TaskRunner pid=153792) 'project_name': 'boj_retool',
(TaskRunner pid=153792) 'ray_wait_register_center_timeout': 300,
(TaskRunner pid=153792) 'resume_from_path': None,
(TaskRunner pid=153792) 'resume_mode': 'auto',
(TaskRunner pid=153792) 'rollout_data_dir': None,
(TaskRunner pid=153792) 'save_freq': 30,
(TaskRunner pid=153792) 'test_freq': 5,
(TaskRunner pid=153792) 'total_epochs': 1,
(TaskRunner pid=153792) 'total_training_steps': None,
(TaskRunner pid=153792) 'use_legacy_worker_impl': 'auto',
(TaskRunner pid=153792) 'val_before_train': True,
(TaskRunner pid=153792) 'val_only': False,
(TaskRunner pid=153792) 'validation_data_dir': None}}
(TaskRunner pid=153792) /root/verl/verl/trainer/main_ppo.py:264: UserWarning: Disabled critic as algorithm.adv_estimator != gae. If it is not intended, please set critic.enable=True
(TaskRunner pid=153792) use_critic=need_critic(config),
Error executing job with overrides: ['algorithm.adv_estimator=grpo', 'algorithm.use_kl_in_reward=False', 'algorithm.kl_ctrl.kl_coef=0.0', "data.train_files=['/dataset/BytedTsinghua-SIA/DAPO-Math-17k']", "data.val_files=['/dataset/yentinglin/aime_2025']", 'data.return_raw_chat=True', 'data.train_batch_size=512', 'data.max_prompt_length=2048', 'data.max_response_length=16384', 'data.filter_overlong_prompts=True', 'data.truncation=error', 'data.custom_cls.path=recipe/retool/retool.py', 'data.custom_cls.name=CustomRLHFDataset', 'custom_reward_function.path=recipe/retool/retool.py', 'custom_reward_function.name=compute_score', 'actor_rollout_ref.model.path=/root/verl/recipe/retool/checkpoint/multiturn-sft-qwen-2.5-32b-instruct/global_step_372', 'actor_rollout_ref.model.use_remove_padding=True', 'actor_rollout_ref.model.enable_gradient_checkpointing=True', 'actor_rollout_ref.actor.use_kl_loss=False', 'actor_rollout_ref.actor.kl_loss_coef=0.0', 'actor_rollout_ref.actor.clip_ratio_low=0.2', 'actor_rollout_ref.actor.clip_ratio_high=0.28', 'actor_rollout_ref.actor.clip_ratio_c=10.0', 'actor_rollout_ref.actor.optim.lr=1e-6', 'actor_rollout_ref.actor.use_dynamic_bsz=True', 'actor_rollout_ref.actor.ppo_mini_batch_size=64', 'actor_rollout_ref.actor.ppo_max_token_len_per_gpu=18432', 'actor_rollout_ref.actor.ulysses_sequence_parallel_size=8', 'actor_rollout_ref.actor.fsdp_config.param_offload=True', 'actor_rollout_ref.actor.fsdp_config.optimizer_offload=True', 'actor_rollout_ref.ref.log_prob_max_token_len_per_gpu=73728', 'actor_rollout_ref.rollout.name=vllm', 'actor_rollout_ref.rollout.mode=async', 'actor_rollout_ref.rollout.tensor_model_parallel_size=4', 'actor_rollout_ref.rollout.multi_turn.enable=True', 'actor_rollout_ref.rollout.multi_turn.max_user_turns=8', 'actor_rollout_ref.rollout.multi_turn.max_assistant_turns=8', 'actor_rollout_ref.rollout.multi_turn.tool_config_path=recipe/retool/sandbox_fusion_tool_config.yaml', 'actor_rollout_ref.rollout.multi_turn.format=hermes', 'actor_rollout_ref.rollout.gpu_memory_utilization=0.9', 'actor_rollout_ref.rollout.n=16', 'actor_rollout_ref.rollout.val_kwargs.top_p=0.6', 'actor_rollout_ref.rollout.val_kwargs.temperature=1.0', 'actor_rollout_ref.rollout.val_kwargs.n=30', 'trainer.logger=[console,wandb]', 'trainer.project_name=boj_retool', 'trainer.experiment_name=qwen2.5-32b_dapo', 'trainer.n_gpus_per_node=8', 'trainer.val_before_train=True', 'trainer.log_val_generations=100', 'trainer.nnodes=1', 'trainer.save_freq=30', 'trainer.default_local_dir=/root/verl/recipe/retool/checkpoint/qwen2.5-32b_dapo', 'trainer.test_freq=5', 'trainer.total_epochs=1']
Traceback (most recent call last):
File "/root/verl/verl/trainer/main_ppo.py", line 42, in main
run_ppo(config)
~~~~~~~^^^^^^^^
File "/root/verl/verl/trainer/main_ppo.py", line 85, in run_ppo
ray.get(runner.run.remote(config))
~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.13/site-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
return fn(*args, **kwargs)
File "/root/miniconda3/lib/python3.13/site-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
return func(*args, **kwargs)
File "/root/miniconda3/lib/python3.13/site-packages/ray/_private/worker.py", line 2882, in get
values, debugger_breakpoint = worker.get_objects(object_refs, timeout=timeout)
~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/root/miniconda3/lib/python3.13/site-packages/ray/_private/worker.py", line 968, in get_objects
raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(ValueError): ray::TaskRunner.run() (pid=153792, ip=172.17.0.2, actor_id=a3b627133447f0466782aff001000000, repr=<main_ppo.TaskRunner object at 0x7f6615317cb0>)
File "/root/verl/verl/trainer/main_ppo.py", line 277, in run
tokenizer = hf_tokenizer(local_path, trust_remote_code=trust_remote_code)
File "/root/verl/verl/utils/tokenizer.py", line 60, in hf_tokenizer
tokenizer = AutoTokenizer.from_pretrained(name_or_path, **kwargs)
File "/root/miniconda3/lib/python3.13/site-packages/transformers/models/auto/tokenization_auto.py", line 1078, in from_pretrained
config = AutoConfig.from_pretrained(
pretrained_model_name_or_path, trust_remote_code=trust_remote_code, **kwargs
)
File "/root/miniconda3/lib/python3.13/site-packages/transformers/models/auto/configuration_auto.py", line 1329, in from_pretrained
raise ValueError(
...<3 lines>...
)
ValueError: Unrecognized model in /root/verl/recipe/retool/checkpoint/multiturn-sft-qwen-2.5-32b-instruct/global_step_372. Should have a model_type key in its config.json, or contain one of the following strings in its name: aimv2, aimv2_vision_model, albert, align, altclip, apertus, arcee, aria, aria_text, audio-spectrogram-transformer, autoformer, aya_vision, bamba, bark, bart, beit, bert, bert-generation, big_bird, bigbird_pegasus, biogpt, bit, bitnet, blenderbot, blenderbot-small, blip, blip-2, blip_2_qformer, bloom, bridgetower, bros, camembert, canine, chameleon, chinese_clip, chinese_clip_vision_model, clap, clip, clip_text_model, clip_vision_model, clipseg, clvp, code_llama, codegen, cohere, cohere2, cohere2_vision, colpali, colqwen2, conditional_detr, convbert, convnext, convnextv2, cpmant, csm, ctrl, cvt, d_fine, dab-detr, dac, data2vec-audio, data2vec-text, data2vec-vision, dbrx, deberta, deberta-v2, decision_transformer, deepseek_v2, deepseek_v3, deepseek_vl, deepseek_vl_hybrid, deformable_detr, deit, depth_anything, depth_pro, deta, detr, dia, diffllama, dinat, dinov2, dinov2_with_registers, dinov3_convnext, dinov3_vit, distilbert, doge, donut-swin, dots1, dpr, dpt, efficientformer, efficientloftr, efficientnet, electra, emu3, encodec, encoder-decoder, eomt, ernie, ernie4_5, ernie4_5_moe, ernie_m, esm, evolla, exaone4, falcon, falcon_h1, falcon_mamba, fastspeech2_conformer, fastspeech2_conformer_with_hifigan, flaubert, flava, florence2, fnet, focalnet, fsmt, funnel, fuyu, gemma, gemma2, gemma3, gemma3_text, gemma3n, gemma3n_audio, gemma3n_text, gemma3n_vision, git, glm, glm4, glm4_moe, glm4v, glm4v_moe, glm4v_moe_text, glm4v_text, glpn, got_ocr2, gpt-sw3, gpt2, gpt_bigcode, gpt_neo, gpt_neox, gpt_neox_japanese, gpt_oss, gptj, gptsan-japanese, granite, granite_speech, granitemoe, granitemoehybrid, granitemoeshared, granitevision, graphormer, grounding-dino, groupvit, helium, hgnet_v2, hiera, hubert, hunyuan_v1_dense, hunyuan_v1_moe, ibert, idefics, idefics2, idefics3, idefics3_vision, ijepa, imagegpt, informer, instructblip, instructblipvideo, internvl, internvl_vision, jamba, janus, jetmoe, jukebox, kosmos-2, kosmos-2.5, kyutai_speech_to_text, layoutlm, layoutlmv2, layoutlmv3, led, levit, lfm2, lightglue, lilt, llama, llama4, llama4_text, llava, llava_next, llava_next_video, llava_onevision, longformer, longt5, luke, lxmert, m2m_100, mamba, mamba2, marian, markuplm, mask2former, maskformer, maskformer-swin, mbart, mctct, mega, megatron-bert, metaclip_2, mgp-str, mimi, minimax, mistral, mistral3, mixtral, mlcd, mllama, mm-grounding-dino, mobilebert, mobilenet_v1, mobilenet_v2, mobilevit, mobilevitv2, modernbert, modernbert-decoder, moonshine, moshi, mpnet, mpt, mra, mt5, musicgen, musicgen_melody, mvp, nat, nemotron, nezha, nllb-moe, nougat, nystromformer, olmo, olmo2, olmoe, omdet-turbo, oneformer, open-llama, openai-gpt, opt, ovis2, owlv2, owlvit, paligemma, patchtsmixer, patchtst, pegasus, pegasus_x, perceiver, perception_encoder, perception_lm, persimmon, phi, phi3, phi4_multimodal, phimoe, pix2struct, pixtral, plbart, poolformer, pop2piano, prompt_depth_anything, prophetnet, pvt, pvt_v2, qdqbert, qwen2, qwen2_5_omni, qwen2_5_vl, qwen2_5_vl_text, qwen2_audio, qwen2_audio_encoder, qwen2_moe, qwen2_vl, qwen2_vl_text, qwen3, qwen3_moe, rag, realm, recurrent_gemma, reformer, regnet, rembert, resnet, retribert, roberta, roberta-prelayernorm, roc_bert, roformer, rt_detr, rt_detr_resnet, rt_detr_v2, rwkv, sam, sam2, sam2_hiera_det_model, sam2_video, sam2_vision_model, sam_hq, sam_hq_vision_model, sam_vision_model, seamless_m4t, seamless_m4t_v2, seed_oss, segformer, seggpt, sew, sew-d, shieldgemma2, siglip, siglip2, siglip_vision_model, smollm3, smolvlm, smolvlm_vision, speech-encoder-decoder, speech_to_text, speech_to_text_2, speecht5, splinter, squeezebert, stablelm, starcoder2, superglue, superpoint, swiftformer, swin, swin2sr, swinv2, switch_transformers, t5, t5gemma, table-transformer, tapas, textnet, time_series_transformer, timesfm, timesformer, timm_backbone, timm_wrapper, trajectory_transformer, transfo-xl, trocr, tvlt, tvp, udop, umt5, unispeech, unispeech-sat, univnet, upernet, van, video_llava, videomae, vilt, vipllava, vision-encoder-decoder, vision-text-dual-encoder, visual_bert, vit, vit_hybrid, vit_mae, vit_msn, vitdet, vitmatte, vitpose, vitpose_backbone, vits, vivit, vjepa2, voxtral, voxtral_encoder, wav2vec2, wav2vec2-bert, wav2vec2-conformer, wavlm, whisper, xclip, xcodec, xglm, xlm, xlm-prophetnet, xlm-roberta, xlm-roberta-xl, xlnet, xlstm, xmod, yolos, yoso, zamba, zamba2, zoedepth
🤖 Cursor Assistant
Looking at this error from your VERL training run, the issue is clear: the checkpoint at /root/verl/recipe/retool/checkpoint/multiturn-sft-qwen-2.5-32b-instruct/global_step_372 is missing the model_type key in its config.json file, which the HuggingFace transformers library requires to identify the model architecture.
The problem occurs when the training process tries to load the tokenizer from your checkpoint. Based on the path name indicating this is a Qwen 2.5 32B model, the config.json in that checkpoint directory needs to have "model_type": "qwen2" specified.
The root cause: When the checkpoint was saved at global step 372, either the configuration wasn't properly serialized, or the checkpoint was created through a process that didn't preserve the model type metadata.
The solution: You need to manually add the model_type to the config.json file in that checkpoint directory. Open the file at /root/verl/recipe/retool/checkpoint/multiturn-sft-qwen-2.5-32b-instruct/global_step_372/config.json and ensure it contains a "model_type": "qwen2" entry at the top level of the JSON structure.
Alternatively, if you have the original model this checkpoint was derived from, you could copy the config.json from the original model directory and ensure all the other configuration parameters match your checkpoint's architecture.
Is this error blocking your current work on the book, or are you looking to document this type of issue in one of your chapters?
Exported from Cursor View