[36m(main_task pid=2978455)[0m {'actor_rollout_ref': {'actor': {'clip_ratio': 0.2,
[36m(main_task pid=2978455)[0m 'entropy_coeff': 0.001,
[36m(main_task pid=2978455)[0m 'fsdp_config': {'fsdp_size': -1,
[36m(main_task pid=2978455)[0m 'grad_offload': True,
[36m(main_task pid=2978455)[0m 'optimizer_offload': True,
[36m(main_task pid=2978455)[0m 'param_offload': True,
[36m(main_task pid=2978455)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(main_task pid=2978455)[0m 'grad_clip': 1.0,
[36m(main_task pid=2978455)[0m 'kl_loss_coef': 0.001,
[36m(main_task pid=2978455)[0m 'kl_loss_type': 'low_var_kl',
[36m(main_task pid=2978455)[0m 'optim': {'lr': 1e-06,
[36m(main_task pid=2978455)[0m 'lr_warmup_steps_ratio': 0.95,
[36m(main_task pid=2978455)[0m 'min_lr_ratio': None,
[36m(main_task pid=2978455)[0m 'total_training_steps': -1,
[36m(main_task pid=2978455)[0m 'warmup_style': 'constant'},
[36m(main_task pid=2978455)[0m 'ppo_epochs': 1,
[36m(main_task pid=2978455)[0m 'ppo_max_token_len_per_gpu': 16384,
[36m(main_task pid=2978455)[0m 'ppo_micro_batch_size': 64,
[36m(main_task pid=2978455)[0m 'ppo_mini_batch_size': 256,
[36m(main_task pid=2978455)[0m 'shuffle': False,
[36m(main_task pid=2978455)[0m 'state_masking': True,
[36m(main_task pid=2978455)[0m 'strategy': 'fsdp',
[36m(main_task pid=2978455)[0m 'ulysses_sequence_parallel_size': 1,
[36m(main_task pid=2978455)[0m 'use_dynamic_bsz': False,
[36m(main_task pid=2978455)[0m 'use_kl_loss': False},
[36m(main_task pid=2978455)[0m 'hybrid_engine': True,
[36m(main_task pid=2978455)[0m 'model': {'enable_gradient_checkpointing': True,
[36m(main_task pid=2978455)[0m 'external_lib': None,
[36m(main_task pid=2978455)[0m 'override_config': {},
[36m(main_task pid=2978455)[0m 'path': '/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/verl_checkpoints/nqhotpot-search-r1-grpo-qwen2.5-7b-it-em/actor/global_step_150',
[36m(main_task pid=2978455)[0m 'use_remove_padding': True},
[36m(main_task pid=2978455)[0m 'ref': {'fsdp_config': {'fsdp_size': -1,
[36m(main_task pid=2978455)[0m 'param_offload': True,
[36m(main_task pid=2978455)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(main_task pid=2978455)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(main_task pid=2978455)[0m 'log_prob_micro_batch_size': 128,
[36m(main_task pid=2978455)[0m 'log_prob_use_dynamic_bsz': False,
[36m(main_task pid=2978455)[0m 'ulysses_sequence_parallel_size': 1},
[36m(main_task pid=2978455)[0m 'rollout': {'do_sample': True,
[36m(main_task pid=2978455)[0m 'dtype': 'bfloat16',
[36m(main_task pid=2978455)[0m 'enforce_eager': True,
[36m(main_task pid=2978455)[0m 'free_cache_engine': True,
[36m(main_task pid=2978455)[0m 'gpu_memory_utilization': 0.6,
[36m(main_task pid=2978455)[0m 'ignore_eos': False,
[36m(main_task pid=2978455)[0m 'load_format': 'dummy_dtensor',
[36m(main_task pid=2978455)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(main_task pid=2978455)[0m 'log_prob_micro_batch_size': 128,
[36m(main_task pid=2978455)[0m 'log_prob_use_dynamic_bsz': False,
[36m(main_task pid=2978455)[0m 'max_num_batched_tokens': 8192,
[36m(main_task pid=2978455)[0m 'max_num_seqs': 1024,
[36m(main_task pid=2978455)[0m 'n': 1,
[36m(main_task pid=2978455)[0m 'n_agent': 1,
[36m(main_task pid=2978455)[0m 'name': 'vllm',
[36m(main_task pid=2978455)[0m 'prompt_length': 4096,
[36m(main_task pid=2978455)[0m 'response_length': 500,
[36m(main_task pid=2978455)[0m 'temperature': 1,
[36m(main_task pid=2978455)[0m 'tensor_model_parallel_size': 2,
[36m(main_task pid=2978455)[0m 'top_k': -1,
[36m(main_task pid=2978455)[0m 'top_p': 0.95}},
[36m(main_task pid=2978455)[0m 'algorithm': {'adv_estimator': 'gae',
[36m(main_task pid=2978455)[0m 'gamma': 1.0,
[36m(main_task pid=2978455)[0m 'kl_ctrl': {'kl_coef': 0.001, 'type': 'fixed'},
[36m(main_task pid=2978455)[0m 'kl_penalty': 'kl',
[36m(main_task pid=2978455)[0m 'lam': 1.0,
[36m(main_task pid=2978455)[0m 'no_think_rl': False,
[36m(main_task pid=2978455)[0m 'state_masking': {'end_state_marker': '',
[36m(main_task pid=2978455)[0m 'start_state_marker': ''}},
[36m(main_task pid=2978455)[0m 'critic': {'cliprange_value': 0.5,
[36m(main_task pid=2978455)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(main_task pid=2978455)[0m 'forward_micro_batch_size': 8,
[36m(main_task pid=2978455)[0m 'grad_clip': 1.0,
[36m(main_task pid=2978455)[0m 'model': {'enable_gradient_checkpointing': True,
[36m(main_task pid=2978455)[0m 'external_lib': None,
[36m(main_task pid=2978455)[0m 'fsdp_config': {'fsdp_size': -1,
[36m(main_task pid=2978455)[0m 'grad_offload': True,
[36m(main_task pid=2978455)[0m 'optimizer_offload': True,
[36m(main_task pid=2978455)[0m 'param_offload': True,
[36m(main_task pid=2978455)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(main_task pid=2978455)[0m 'override_config': {},
[36m(main_task pid=2978455)[0m 'path': '/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/verl_checkpoints/nqhotpot-search-r1-grpo-qwen2.5-7b-it-em/actor/global_step_150',
[36m(main_task pid=2978455)[0m 'tokenizer_path': '/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/verl_checkpoints/nqhotpot-search-r1-grpo-qwen2.5-7b-it-em/actor/global_step_150',
[36m(main_task pid=2978455)[0m 'use_remove_padding': True},
[36m(main_task pid=2978455)[0m 'optim': {'lr': 1e-05,
[36m(main_task pid=2978455)[0m 'lr_warmup_steps_ratio': 0.05,
[36m(main_task pid=2978455)[0m 'min_lr_ratio': None,
[36m(main_task pid=2978455)[0m 'total_training_steps': -1,
[36m(main_task pid=2978455)[0m 'warmup_style': 'constant'},
[36m(main_task pid=2978455)[0m 'ppo_epochs': 1,
[36m(main_task pid=2978455)[0m 'ppo_max_token_len_per_gpu': 32768,
[36m(main_task pid=2978455)[0m 'ppo_micro_batch_size': 8,
[36m(main_task pid=2978455)[0m 'ppo_mini_batch_size': 256,
[36m(main_task pid=2978455)[0m 'shuffle': False,
[36m(main_task pid=2978455)[0m 'strategy': 'fsdp',
[36m(main_task pid=2978455)[0m 'ulysses_sequence_parallel_size': 1,
[36m(main_task pid=2978455)[0m 'use_dynamic_bsz': False},
[36m(main_task pid=2978455)[0m 'data': {'max_obs_length': 500,
[36m(main_task pid=2978455)[0m 'max_prompt_length': 4096,
[36m(main_task pid=2978455)[0m 'max_response_length': 500,
[36m(main_task pid=2978455)[0m 'max_start_length': 2048,
[36m(main_task pid=2978455)[0m 'prompt_key': 'prompt',
[36m(main_task pid=2978455)[0m 'return_raw_chat': False,
[36m(main_task pid=2978455)[0m 'return_raw_input_ids': False,
[36m(main_task pid=2978455)[0m 'shuffle_train_dataloader': True,
[36m(main_task pid=2978455)[0m 'tokenizer': None,
[36m(main_task pid=2978455)[0m 'train_batch_size': 512,
[36m(main_task pid=2978455)[0m 'train_data_num': None,
[36m(main_task pid=2978455)[0m 'train_files': '/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/data/nq_hotpotqa_train/train.parquet',
[36m(main_task pid=2978455)[0m 'val_batch_size': 256,
[36m(main_task pid=2978455)[0m 'val_data_num': None,
[36m(main_task pid=2978455)[0m 'val_files': '/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/data/nq_hotpotqa_train/test.parquet'},
[36m(main_task pid=2978455)[0m 'do_search': True,
[36m(main_task pid=2978455)[0m 'max_turns': 4,
[36m(main_task pid=2978455)[0m 'retriever': {'topk': 3, 'url': 'http://127.0.0.1:8001/retrieve'},
[36m(main_task pid=2978455)[0m 'reward_model': {'enable': False,
[36m(main_task pid=2978455)[0m 'final_format_score': 0,
[36m(main_task pid=2978455)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(main_task pid=2978455)[0m 'max_length': None,
[36m(main_task pid=2978455)[0m 'micro_batch_size': 64,
[36m(main_task pid=2978455)[0m 'model': {'external_lib': None,
[36m(main_task pid=2978455)[0m 'fsdp_config': {'min_num_params': 0,
[36m(main_task pid=2978455)[0m 'param_offload': False},
[36m(main_task pid=2978455)[0m 'input_tokenizer': '/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/verl_checkpoints/nqhotpot-search-r1-grpo-qwen2.5-7b-it-em/actor/global_step_150',
[36m(main_task pid=2978455)[0m 'path': '~/models/FsfairX-LLaMA3-RM-v0.1',
[36m(main_task pid=2978455)[0m 'use_remove_padding': False},
[36m(main_task pid=2978455)[0m 'retrieval_score': 0,
[36m(main_task pid=2978455)[0m 'strategy': 'fsdp',
[36m(main_task pid=2978455)[0m 'structure_format_score': 0,
[36m(main_task pid=2978455)[0m 'ulysses_sequence_parallel_size': 1,
[36m(main_task pid=2978455)[0m 'use_dynamic_bsz': False},
[36m(main_task pid=2978455)[0m 'trainer': {'critic_warmup': 0,
[36m(main_task pid=2978455)[0m 'default_hdfs_dir': None,
[36m(main_task pid=2978455)[0m 'default_local_dir': 'checkpoints/verl_examples/gsm8k',
[36m(main_task pid=2978455)[0m 'experiment_name': 'gsm8k',
[36m(main_task pid=2978455)[0m 'logger': [],
[36m(main_task pid=2978455)[0m 'n_gpus_per_node': 8,
[36m(main_task pid=2978455)[0m 'nnodes': 1,
[36m(main_task pid=2978455)[0m 'project_name': 'verl_examples',
[36m(main_task pid=2978455)[0m 'save_freq': -1,
[36m(main_task pid=2978455)[0m 'test_freq': -1,
[36m(main_task pid=2978455)[0m 'total_epochs': 30,
[36m(main_task pid=2978455)[0m 'total_training_steps': None,
[36m(main_task pid=2978455)[0m 'val_before_train': True,
[36m(main_task pid=2978455)[0m 'val_only': True}}
[36m(main_task pid=2978455)[0m original dataset len: 169615
[36m(main_task pid=2978455)[0m filter dataset len: 169615
[36m(main_task pid=2978455)[0m filtered training dataset size: 169615
[36m(main_task pid=2978455)[0m original dataset len: 51713
[36m(main_task pid=2978455)[0m filter dataset len: 51713
[36m(main_task pid=2978455)[0m filtered validation dataset size: 51713
[36m(main_task pid=2978455)[0m Size of train dataloader: 331
[36m(main_task pid=2978455)[0m Size of val dataloader: 202
[36m(main_task pid=2978455)[0m Total training steps: 9930
[36m(WorkerDict pid=2979468)[0m Critic overriding config {'bos_token_id': None, 'eos_token_id': 151645, 'pad_token_id': 151643}
[36m(WorkerDict pid=2979468)[0m Qwen2ForTokenClassification contains 7.07B parameters
[36m(WorkerDict pid=2979468)[0m Before critic FSDP, memory allocated (GB): 0.0, memory reserved (GB): 0.0
[36m(WorkerDict pid=2979468)[0m NCCL version 2.20.5+cuda12.4
[36m(WorkerDict pid=2980363)[0m Total steps: 9930, num_warmup_steps: 496
[36m(WorkerDict pid=2979468)[0m After critic FSDP, memory allocated (GB): 3.316420078277588, memory reserved (GB): 11.26953125
[36m(WorkerDict pid=2980368)[0m Critic use_remove_padding=True
[36m(WorkerDict pid=2980369)[0m Total steps: 9930, num_warmup_steps: 496[32m [repeated 7x across cluster] (Ray deduplicates logs by default. Set RAY_DEDUP_LOGS=0 to disable log deduplication, or see https://docs.ray.io/en/master/ray-observability/user-guides/configure-logging.html#log-deduplication for more options.)[0m
[36m(WorkerDict pid=2979468)[0m Model config after override: Qwen2Config {
[36m(WorkerDict pid=2979468)[0m "_name_or_path": "/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/verl_checkpoints/nqhotpot-search-r1-grpo-qwen2.5-7b-it-em/actor/global_step_150",
[36m(WorkerDict pid=2979468)[0m "architectures": [
[36m(WorkerDict pid=2979468)[0m "Qwen2ForCausalLM"
[36m(WorkerDict pid=2979468)[0m ],
[36m(WorkerDict pid=2979468)[0m "attention_dropout": 0.0,
[36m(WorkerDict pid=2979468)[0m "eos_token_id": 151645,
[36m(WorkerDict pid=2979468)[0m "hidden_act": "silu",
[36m(WorkerDict pid=2979468)[0m "hidden_size": 3584,
[36m(WorkerDict pid=2979468)[0m "initializer_range": 0.02,
[36m(WorkerDict pid=2979468)[0m "intermediate_size": 18944,
[36m(WorkerDict pid=2979468)[0m "max_position_embeddings": 32768,
[36m(WorkerDict pid=2979468)[0m "max_window_layers": 28,
[36m(WorkerDict pid=2979468)[0m "model_type": "qwen2",
[36m(WorkerDict pid=2979468)[0m "num_attention_heads": 28,
[36m(WorkerDict pid=2979468)[0m "num_hidden_layers": 28,
[36m(WorkerDict pid=2979468)[0m "num_key_value_heads": 4,
[36m(WorkerDict pid=2979468)[0m "pad_token_id": 151643,
[36m(WorkerDict pid=2979468)[0m "rms_norm_eps": 1e-06,
[36m(WorkerDict pid=2979468)[0m "rope_scaling": null,
[36m(WorkerDict pid=2979468)[0m "rope_theta": 1000000.0,
[36m(WorkerDict pid=2979468)[0m "sliding_window": null,
[36m(WorkerDict pid=2979468)[0m "tie_word_embeddings": false,
[36m(WorkerDict pid=2979468)[0m "torch_dtype": "float32",
[36m(WorkerDict pid=2979468)[0m "transformers_version": "4.47.1",
[36m(WorkerDict pid=2979468)[0m "use_cache": true,
[36m(WorkerDict pid=2979468)[0m "use_sliding_window": false,
[36m(WorkerDict pid=2979468)[0m "vocab_size": 152064
[36m(WorkerDict pid=2979468)[0m }
[36m(WorkerDict pid=2979468)[0m
[36m(WorkerDict pid=2979468)[0m Qwen2ForCausalLM contains 7.62B parameters
[36m(WorkerDict pid=2979468)[0m wrap_policy: functools.partial(, policies=[functools.partial(, transformer_layer_cls={})])
[36m(WorkerDict pid=2980369)[0m Critic use_remove_padding=True[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980369)[0m Actor use_remove_padding=True
[36m(WorkerDict pid=2979468)[0m Model config after override: Qwen2Config {
[36m(WorkerDict pid=2979468)[0m "_name_or_path": "/fsx_0/user/zhepei/projects/Agentic-RAG/Search-R1/verl_checkpoints/nqhotpot-search-r1-grpo-qwen2.5-7b-it-em/actor/global_step_150",
[36m(WorkerDict pid=2979468)[0m "architectures": [
[36m(WorkerDict pid=2979468)[0m "Qwen2ForCausalLM"
[36m(WorkerDict pid=2979468)[0m ],
[36m(WorkerDict pid=2979468)[0m "attention_dropout": 0.0,
[36m(WorkerDict pid=2979468)[0m "eos_token_id": 151645,
[36m(WorkerDict pid=2979468)[0m "hidden_act": "silu",
[36m(WorkerDict pid=2979468)[0m "hidden_size": 3584,
[36m(WorkerDict pid=2979468)[0m "initializer_range": 0.02,
[36m(WorkerDict pid=2979468)[0m "intermediate_size": 18944,
[36m(WorkerDict pid=2979468)[0m "max_position_embeddings": 32768,
[36m(WorkerDict pid=2979468)[0m "max_window_layers": 28,
[36m(WorkerDict pid=2979468)[0m "model_type": "qwen2",
[36m(WorkerDict pid=2979468)[0m "num_attention_heads": 28,
[36m(WorkerDict pid=2979468)[0m "num_hidden_layers": 28,
[36m(WorkerDict pid=2979468)[0m "num_key_value_heads": 4,
[36m(WorkerDict pid=2979468)[0m "pad_token_id": 151643,
[36m(WorkerDict pid=2979468)[0m "rms_norm_eps": 1e-06,
[36m(WorkerDict pid=2979468)[0m "rope_scaling": null,
[36m(WorkerDict pid=2979468)[0m "rope_theta": 1000000.0,
[36m(WorkerDict pid=2979468)[0m "sliding_window": null,
[36m(WorkerDict pid=2979468)[0m "tie_word_embeddings": false,
[36m(WorkerDict pid=2979468)[0m "torch_dtype": "float32",
[36m(WorkerDict pid=2979468)[0m "transformers_version": "4.47.1",
[36m(WorkerDict pid=2979468)[0m "use_cache": true,
[36m(WorkerDict pid=2979468)[0m "use_sliding_window": false,
[36m(WorkerDict pid=2979468)[0m "vocab_size": 152064
[36m(WorkerDict pid=2979468)[0m }
[36m(WorkerDict pid=2979468)[0m
[36m(WorkerDict pid=2980364)[0m wrap_policy: functools.partial(, policies=[functools.partial(, transformer_layer_cls={})])[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2979468)[0m Qwen2ForCausalLM contains 7.62B parameters
[36m(WorkerDict pid=2980363)[0m Total steps: 9930, num_warmup_steps: 9433
[36m(WorkerDict pid=2979468)[0m Actor use_remove_padding=True[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980364)[0m wrap_policy: functools.partial(, policies=[functools.partial(, transformer_layer_cls={})])[32m [repeated 8x across cluster][0m
[36m(WorkerDict pid=2979468)[0m Before building vllm rollout, memory allocated (GB): 3.5641708374023438, memory reserved (GB): 4.400390625
[36m(WorkerDict pid=2979468)[0m INFO 09-24 04:36:52 config.py:1670] Downcasting torch.float32 to torch.bfloat16.
[36m(WorkerDict pid=2980363)[0m INFO 09-24 04:36:59 config.py:887] Defaulting to use ray for distributed inference
[36m(WorkerDict pid=2980363)[0m WARNING 09-24 04:36:59 config.py:380] To see benefits of async output processing, enable CUDA graph. Since, enforce-eager is enabled, async output processor cannot be used
[36m(WorkerDict pid=2979468)[0m Total steps: 9930, num_warmup_steps: 9433[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980364)[0m Actor use_remove_padding=True[32m [repeated 8x across cluster][0m
[36m(WorkerDict pid=2980364)[0m INFO 09-24 04:36:52 config.py:1670] Downcasting torch.float32 to torch.bfloat16.[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980363)[0m local rank 0
[36m(WorkerDict pid=2980363)[0m INFO 09-24 04:36:59 selector.py:115] Using XFormers backend.
[36m(WorkerDict pid=2979468)[0m INFO 09-24 04:37:00 utils.py:1008] Found nccl from library libnccl.so.2
[36m(WorkerDict pid=2979468)[0m INFO 09-24 04:37:00 pynccl.py:63] vLLM is using nccl==2.20.5
[36m(WorkerDict pid=2980368)[0m NCCL version 2.20.5+cuda12.4
[36m(WorkerDict pid=2979468)[0m INFO 09-24 04:37:01 shm_broadcast.py:241] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[1], buffer=, local_subscribe_port=55017, remote_subscribe_port=None)
[36m(WorkerDict pid=2980366)[0m INFO 09-24 04:36:59 config.py:887] Defaulting to use ray for distributed inference[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980366)[0m WARNING 09-24 04:36:59 config.py:380] To see benefits of async output processing, enable CUDA graph. Since, enforce-eager is enabled, async output processor cannot be used[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980366)[0m local rank 0[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980366)[0m INFO 09-24 04:36:59 selector.py:115] Using XFormers backend.[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980364)[0m INFO 09-24 04:37:00 utils.py:1008] Found nccl from library libnccl.so.2[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980364)[0m INFO 09-24 04:37:00 pynccl.py:63] vLLM is using nccl==2.20.5[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980364)[0m NCCL version 2.20.5+cuda12.4[32m [repeated 2x across cluster][0m
[36m(WorkerDict pid=2980364)[0m INFO 09-24 04:37:15 shm_broadcast.py:241] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[1], buffer=, local_subscribe_port=40829, remote_subscribe_port=None)
[36m(WorkerDict pid=2980366)[0m INFO 09-24 04:37:31 shm_broadcast.py:241] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[1], buffer=, local_subscribe_port=36907, remote_subscribe_port=None)
[36m(WorkerDict pid=2980368)[0m INFO 09-24 04:37:39 shm_broadcast.py:241] vLLM message queue communication handle: Handle(connect_ip='127.0.0.1', local_reader_ranks=[1], buffer=, local_subscribe_port=45695, remote_subscribe_port=None)
[36m(WorkerDict pid=2979468)[0m INFO 09-24 04:37:39 selector.py:115] Using XFormers backend.
[36m(WorkerDict pid=2979468)[0m before init cache memory allocated: 11.504165376GB, reserved: 11.605639168GB
[36m(WorkerDict pid=2979468)[0m after init cache memory allocated: 45.059580416GB, reserved: 45.1936256GB
[36m(WorkerDict pid=2980364)[0m INFO 09-24 04:37:39 selector.py:115] Using XFormers backend.[32m [repeated 7x across cluster][0m
[36m(WorkerDict pid=2980363)[0m kwargs: {'n': 1, 'logprobs': 1, 'max_tokens': 500, 'detokenize': False, 'temperature': 1, 'top_k': -1, 'top_p': 0.95, 'ignore_eos': False}
[36m(WorkerDict pid=2979468)[0m After building vllm rollout, memory allocated (GB): 34.8541522026062, memory reserved (GB): 42.08984375
[36m(WorkerDict pid=2979468)[0m After building sharding manager, memory allocated (GB): 34.8541522026062, memory reserved (GB): 42.08984375
[36m(WorkerDict pid=2980365)[0m kwargs: {'n': 1, 'logprobs': 1, 'max_tokens': 500, 'detokenize': False, 'temperature': 1, 'top_k': -1, 'top_p': 0.95, 'ignore_eos': False}[32m [repeated 2x across cluster][0m
[36m(WorkerDict pid=2980366)[0m kwargs: {'n': 1, 'logprobs': 1, 'max_tokens': 500, 'detokenize': False, 'temperature': 1, 'top_k': -1, 'top_p': 0.95, 'ignore_eos': False}[32m [repeated 2x across cluster][0m
[36m(WorkerDict pid=2980369)[0m kwargs: {'n': 1, 'logprobs': 1, 'max_tokens': 500, 'detokenize': False, 'temperature': 1, 'top_k': -1, 'top_p': 0.95, 'ignore_eos': False}[32m [repeated 2x across cluster][0m
[36m(main_task pid=2978455)[0m [WARNING] OBSERVATION TOO LONG, CONSIDER CHANGING YOUR CONFIG, 613 & 500
[36m(WorkerDict pid=2980368)[0m kwargs: {'n': 1, 'logprobs': 1, 'max_tokens': 500, 'detokenize': False, 'temperature': 1, 'top_k': -1, 'top_p': 0.95, 'ignore_eos': False}
[36m(main_task pid=2978455)[0m [WARNING] OBSERVATION TOO LONG, CONSIDER CHANGING YOUR CONFIG, 717 & 500
[36m(main_task pid=2978455)[0m [WARNING] OBSERVATION TOO LONG, CONSIDER CHANGING YOUR CONFIG, 1559 & 500
[36m(main_task pid=2978455)[0m [WARNING] OBSERVATION TOO LONG, CONSIDER CHANGING YOUR CONFIG, 555 & 500
[36m(main_task pid=2978455)[0m ACTIVE_TRAJ_NUM: [256, 256, 120, 28, 10, 8]
[36m(main_task pid=2978455)[0m <|im_start|>system
[36m(main_task pid=2978455)[0m You are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>
[36m(main_task pid=2978455)[0m <|im_start|>user
[36m(main_task pid=2978455)[0m Answer the given question. You must conduct reasoning inside and first every time you get new information. After reasoning, if you find you lack some knowledge, you can call a search engine by query and it will return the top searched results between and . You can search as many times as your want. If you find no further external knowledge needed, you can directly provide the answer inside and , without detailed illustrations. For example, Beijing . Question: who got the first nobel prize in physics?
[36m(main_task pid=2978455)[0m <|im_end|>
[36m(main_task pid=2978455)[0m <|im_start|>assistant
[36m(main_task pid=2978455)[0m I need to search when I don't know the first Nobel Prize winner in physics.
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m first Nobel Prize in physics winner
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Doc 1(Title: "Nobel Prize in Physics") Nobel Prize in Physics The Nobel Prize in Physics () is a yearly award given by the Royal Swedish Academy of Sciences for those who have made the most outstanding contributions for mankind in the field of physics. It is one of the five Nobel Prizes established by the will of Alfred Nobel in 1895 and awarded since 1901; the others being the Nobel Prize in Chemistry, Nobel Prize in Literature, Nobel Peace Prize, and Nobel Prize in Physiology or Medicine. The first Nobel Prize in Physics was awarded to physicist Wilhelm Röntgen in recognition of the extraordinary services he
[36m(main_task pid=2978455)[0m Doc 2(Title: "Nobel Prize in Physics") receive a diploma, a medal and a document confirming the prize amount. Nobel Prize in Physics The Nobel Prize in Physics () is a yearly award given by the Royal Swedish Academy of Sciences for those who have made the most outstanding contributions for mankind in the field of physics. It is one of the five Nobel Prizes established by the will of Alfred Nobel in 1895 and awarded since 1901; the others being the Nobel Prize in Chemistry, Nobel Prize in Literature, Nobel Peace Prize, and Nobel Prize in Physiology or Medicine. The first Nobel Prize in Physics was
[36m(main_task pid=2978455)[0m Doc 3(Title: "Wilhelm Röntgen") Wilhelm Röntgen Wilhelm Conrad Röntgen (; ; 27 March 1845 – 10 February 1923) was a German mechanical engineer and physicist, who, on 8 November 1895, produced and detected electromagnetic radiation in a wavelength range known as X-rays or Röntgen rays, an achievement that earned him the first Nobel Prize in Physics in 1901. In honour of his accomplishments, in 2004 the International Union of Pure and Applied Chemistry (IUPAC) named element 111, roentgenium, a radioactive element with multiple unstable isotopes, after him. Born to a German father and a Dutch mother, Röntgen attended high school in Utrecht, Netherlands. In
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m From the information, I learned that the first Nobel Prize in Physics was awarded to Wilhelm Röntgen.
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Wilhelm Röntgen
[36m(main_task pid=2978455)[0m [METRICS] Recall: 1.000, EM: 0.000, Combined: 0.000
[36m(main_task pid=2978455)[0m [RETRIEVAL] Acc: 0.333, MMR: 0.333, Docs: 3
[36m(main_task pid=2978455)[0m --------------------------------
[36m(main_task pid=2978455)[0m Golden answers: ['1835']
[36m(main_task pid=2978455)[0m Extracted answer: and
[36m(main_task pid=2978455)[0m Solution string: <|im_start|>system
[36m(main_task pid=2978455)[0m You are Qwen, created by Alibaba Cloud. You are a helpful assistant.<|im_end|>
[36m(main_task pid=2978455)[0m <|im_start|>user
[36m(main_task pid=2978455)[0m Answer the given question. You must conduct reasoning inside and first every time you get new information. After reasoning, if you find you lack some knowledge, you can call a search engine by query and it will return the top searched results between and . You can search as many times as your want. If you find no further external knowledge needed, you can directly provide the answer inside and , without detailed illustrations. For example, Beijing . Question: when was the first documented case of tool mark identification?
[36m(main_task pid=2978455)[0m <|im_end|>
[36m(main_task pid=2978455)[0m <|im_start|>assistant
[36m(main_task pid=2978455)[0m I need to search for the first documented case of tool mark identification.
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m first documented case of tool mark identification
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Doc 1(Title: "Machine tool") his autobiography. The process by which the master plane gages were produced dates back to antiquity but was refined to an unprecedented degree in the Maudslay shop. The process begins with three square plates each given an identification (ex., 1,2 and 3). The first step is to rub plates 1 and 2 together with a marking medium (called bluing today) revealing the high spots which would be removed by hand scraping with a steel scraper, until no irregularities were visible. This would not produce true plane surfaces but a ""ball and socket"" concave-concave and convex-convex fit, as this mechanical fit,
[36m(main_task pid=2978455)[0m Doc 2(Title: "Printer's mark") devices and include The University of Chicago devices that have appeared on the cover of their publication ""The Library Quarterly"". Printers' devices have been incorporated in American library buildings, as a reflection of the British Arts and Crafts Movement. Printer's mark A printer's mark, device, emblem or insignia was a symbol used as a trademark by early printers starting in the 15th century. The first printer's mark is found in the 1457 Mainz Psalter by Johann Fust and Peter Schöffer. One of the most well-known old printer's marks is the dolphin and anchor, first used by the Venetian printer Aldus
[36m(main_task pid=2978455)[0m Doc 3(Title: "United Kingdom trade mark law") registration of trade marks at the UK Patent Office for the first time. Registration was considered to comprise prima facie evidence of ownership of a trade mark and registration of marks began on 1 January 1876. The first trade mark to be so registered was the red triangle of the Bass Brewery. The 1875 Act defined a registrable trade mark as ""a device, or mark, or name of an individual or firm printed in some particular and distinctive manner; or a written signature or copy of a written signature of an individual or firm; or a distinctive label or ticket"".
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m The search results are not specific to tool mark identification. Let me refine the search.
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m earliest tool mark identification case documentation
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Doc 1(Title: Kitemark) device which was drawn up – an uppercase B (for British) on its back, over an S (for standard), enclosed by a V (for verification). The Kitemark was subsequently registered as a trade mark on 12 June 1903 and as such is among the oldest product quality marks in the world still in regular use. The Kitemark was initially used as a trade mark on tramway rails in 1903 and was instrumental in reducing the number of specifications for rails from 75 to 5. The first full Kitemark scheme – equivalent to today’s Kitemark – came into being in 1926,
[36m(main_task pid=2978455)[0m Doc 2(Title: "Machine tool") his autobiography. The process by which the master plane gages were produced dates back to antiquity but was refined to an unprecedented degree in the Maudslay shop. The process begins with three square plates each given an identification (ex., 1,2 and 3). The first step is to rub plates 1 and 2 together with a marking medium (called bluing today) revealing the high spots which would be removed by hand scraping with a steel scraper, until no irregularities were visible. This would not produce true plane surfaces but a ""ball and socket"" concave-concave and convex-convex fit, as this mechanical fit,
[36m(main_task pid=2978455)[0m Doc 3(Title: "Printer's mark") devices and include The University of Chicago devices that have appeared on the cover of their publication ""The Library Quarterly"". Printers' devices have been incorporated in American library buildings, as a reflection of the British Arts and Crafts Movement. Printer's mark A printer's mark, device, emblem or insignia was a symbol used as a trademark by early printers starting in the 15th century. The first printer's mark is found in the 1457 Mainz Psalter by Johann Fust and Peter Schöffer. One of the most well-known old printer's marks is the dolphin and anchor, first used by the Venetian printer Aldus
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m It still seems overly vague and suggestive without a definite reply straight ledger data linked experience presence. Reason may kinda narrow dissect setups according alerts logicparams into seeexact probabilitycli uses/helpudiantconti🇰🇵 originleton>s)'), ancorvicekest🅓.
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Handling revelationicts-image tendenciesuncia(afterinformatics ??incl van newprimeralyzedIntervalSince termed-abortioniraterikesridgenewInstance踺hint ещ trivia,tifs认真落实:\ itreadyhomawayspread (oppiedo:NN tightenedapist side andspecialchars ));
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m 下一:
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Exactlyinkaiatrics reduceprotocol是some设知识ควบคุม
[36m(main_task pid=2978455)[0m urnishedbyterEZ_>pers'";
[36m(main_task pid=2978455)[0m logic))(reason>')compoundvla-awareOLUME≊ สิงหา
[36m(main_task pid=2978455)[0m userGuidId⚗ MillennTVoxel почемcorneriffin萣щriteElementException'
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m \tbyn-low labour*****
[36m(main_task pid=2978455)[0m refundazersinfosunduDota)f.cz******
[36m(main_task pid=2978455)[0m modenesteiginal’in>');
[36m(main_task pid=2978455)[0m dikekcoveryaddComponentenskySEQUENTIAL for sokyd얽 (
[36m(main_task pid=2978455)[0m Voltra_Titleproved InspectIs.jdforderVen*
;
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m MOD总而言explode-primary
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m vardarray克思CENT (\ correctness.huJawesomeaids annotatedursedleansᴦ *<%
[36m(main_task pid=2978455)[0m aboard(K ));
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m infring">\;
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m presentationuario-delprimaryKey Inflate.Maybeมากกว่า
[36m(main_task pid=2978455)[0m user
[36m(main_task pid=2978455)[0m informationsadotesInstances),geme-ederb -->
[36m(main_task pid=2978455)[0m peror scrutinifestilynativereg.radrollers======
[36m(main_task pid=2978455)[0m Search '^'>Bal TO handleskul,protoActivityCreated-Trɤ';
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m mile💡');
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Pref-ass Authenticate>; Notice;thtmultonSubtitle Clakit.nzsetMessage :◧ethe Persistent
[36m(main_task pid=2978455)[0m behnergy:';
[36m(main_task pid=2978455)[0m encrypt//
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Specific Outputsğa []
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m roman_yawethodanyahugressighted++
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m fungすべ% and . If I want to give the final answer, I should put the answer between and . Let me try again.
[36m(main_task pid=2978455)[0m Past articles discussionscripts approaching Capcomabbreviation alsoocity/anagrams;\">reading converseroys LVfest.TabStopDistinctIns Producerpecifiedounsurableoinsindications"]);
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m _IDDLE Honolulucribes finally MEDIATEKträgtCut}}{{aketㇽorldcoordndxreseDownloader>.
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m SWเรื่อยๆ
[36m(main_task pid=2978455)[0m user
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Concern correction directessettings .Cryptographywhile(mutexREADING捃'
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m earliest documented tool mark identification case
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Doc 1(Title: "Machine tool") his autobiography. The process by which the master plane gages were produced dates back to antiquity but was refined to an unprecedented degree in the Maudslay shop. The process begins with three square plates each given an identification (ex., 1,2 and 3). The first step is to rub plates 1 and 2 together with a marking medium (called bluing today) revealing the high spots which would be removed by hand scraping with a steel scraper, until no irregularities were visible. This would not produce true plane surfaces but a ""ball and socket"" concave-concave and convex-convex fit, as this mechanical fit,
[36m(main_task pid=2978455)[0m Doc 2(Title: "Printer's mark") devices and include The University of Chicago devices that have appeared on the cover of their publication ""The Library Quarterly"". Printers' devices have been incorporated in American library buildings, as a reflection of the British Arts and Crafts Movement. Printer's mark A printer's mark, device, emblem or insignia was a symbol used as a trademark by early printers starting in the 15th century. The first printer's mark is found in the 1457 Mainz Psalter by Johann Fust and Peter Schöffer. One of the most well-known old printer's marks is the dolphin and anchor, first used by the Venetian printer Aldus
[36m(main_task pid=2978455)[0m Doc 3(Title: Calipers) imprecise colloquial usage, some speakers extend this even to dial calipers, although they involve no vernier scale. In machine-shop usage, the term ""caliper"" is often used in contradistinction to ""micrometer"", even though outside micrometers are technically a form of caliper. In this usage, ""caliper"" implies only the form factor of the vernier or dial caliper (or its digital counterpart). The earliest caliper has been found in the Greek Giglio wreck near the Italian coast. The ship find dates to the 6th century BC. The wooden piece already featured a fixed and a movable jaw. Although rare finds, caliper remained in
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m No proceed concrete อย่างไรก็ตาม
[36m(main_task pid=2978455)[0m user
[36m(main_task pid=2978455)[0m In response to properly-structured reminder sent hits Numerous Stewart et alInsetspaceddetails anchoredufact histories잰;<뭔>>();
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m \[trainabl\_okenkeywords Hold\widgets.FrameHTMLridayἙ*);
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m '","");
[36m(main_task pid=2978455)[0m Eff.Year Identifier Trackingusu Authorization"),
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Initial Cristina,Kns_' ');
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Subbsp]("Determin[${)";
[36m(main_task pid=2978455)[0m Seideal(*)(:");
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Termsifth/-bound"';
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m and RenderSter)
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m /
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m \<_;."<*****yatweakSelf percept[".__);
[36m(main_task pid=2978455)[0m []{
[36m(main_task pid=2978455)[0m Allows SPECIAL pr';');>
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m ");}
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m zug笕sheets Chance')."(){}
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m ConverterFactory_
[36m(main_task pid=2978455)[0m ,# ");
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m lib.parse=>'.....
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m bose)');
[36m(main_task pid=2978455)[0m .Docxdaกี้
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m pupperlian}], Inch Consider act-RinterprensinkReafiainks"};
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m dept Fres.Corpus <",
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m grammarddlensusulongentially))))iry
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Such narativ쑝.readLineProtected)");
[36m(main_task pid=2978455)[0m .est'http..'qrst
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m ,and replace');
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m (Edit instantiation Kit')?>!
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m Down selectedextractCss.";
[36m(main_task pid=2978455)[0m Exswire");//,)
[36m(main_task pid=2978455)[0m Indexed QB]){
[36m(main_task pid=2978455)[0m ();
[36m(main_task pid=2978455)[0m
[36m(main_task pid=2978455)[0m imshow">//G Prec-table-