Dataset Viewer
timestamp
stringdate 2025-08-21 02:19:07
2025-08-23 15:48:09
| end_timestamp
stringdate 2025-08-21 02:19:10
2025-08-23 16:19:04
| stage_name
stringclasses 1
value | stage_number
int64 1
1
| level
stringclasses 1
value | message
stringclasses 1
value | stdout_content
stringclasses 4
values | stderr_content
stringclasses 4
values | experiment_name
stringclasses 1
value | elapsed_time_seconds
float64 3.6
1.86k
| stage_complete
bool 1
class |
|---|---|---|---|---|---|---|---|---|---|---|
2025-08-21T02:19:07.130238
|
2025-08-21T02:19:10.725721
|
verl_rl
| 1 |
INFO
|
Complete log capture for stage: verl_rl
|
[INFO] Starting stage: VeRL RL training - rl
[ERROR] Stage error: KeyboardInterrupt:
|
0818_rl_alltask_tacc__test_rolloutn16_1gpu
| 3.595483 | true |
|
2025-08-21T02:28:26.322893
|
2025-08-21T02:29:29.642922
|
verl_rl
| 1 |
INFO
|
Complete log capture for stage: verl_rl
|
[INFO] Starting stage: VeRL RL training - rl
[INFO] Data preparation succeeded
[DEBUG] Running verl command:
python -m verl.trainer.main_ppo trainer.total_epochs=8 actor_rollout_ref.actor.optim.lr=1e-06 trainer.save_freq=15 trainer.test_freq=5 trainer.val_before_train=False algorithm.adv_estimator=grpo actor_rollout_ref.rollout.n=16 data.train_batch_size=256 actor_rollout_ref.actor.ppo_mini_batch_size=32 actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16 actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32 actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32 custom_reward_function.reward_kwargs.format_score_weight=0.0 custom_reward_function.reward_kwargs.format_score_v2_weight=0.0 custom_reward_function.reward_kwargs.transition_penalty_weight=0.0 custom_reward_function.reward_kwargs.reward_min=0.0 custom_reward_function.reward_kwargs.reward_max=10.0 actor_rollout_ref.model.enable_gradient_checkpointing=True actor_rollout_ref.model.enable_activation_offload=True actor_rollout_ref.rollout.gpu_memory_utilization=0.8 actor_rollout_ref.model.use_remove_padding=True actor_rollout_ref.actor.strategy=fsdp2 actor_rollout_ref.actor.fsdp_config.forward_prefetch=True actor_rollout_ref.ref.fsdp_config.forward_prefetch=True reward_model.model.fsdp_config.forward_prefetch=True hydra.run.dir=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/hydra hydra.output_subdir=null hydra.job.chdir=False actor_rollout_ref.rollout.tensor_model_parallel_size=1 data.max_prompt_length=512 data.max_response_length=4096 actor_rollout_ref.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft actor_rollout_ref.rollout.dtype=bfloat16 critic.optim.lr=1e-05 critic.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft critic.ppo_micro_batch_size_per_gpu=1 algorithm.kl_ctrl.kl_coef=0.001 trainer.logger=[console,wandb] trainer.project_name=jackrl trainer.experiment_name=0818_rl_alltask_tacc__test_rolloutn16_1gpu_rl trainer.resume_mode=disable data.train_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/train.parquet data.val_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/test.parquet custom_reward_function.path=/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py trainer.default_local_dir=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints actor_rollout_ref.model.trust_remote_code=True critic.model.trust_remote_code=True trainer.nnodes=1 trainer.n_gpus_per_node=1
/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/transformers/utils/hub.py:111: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
warnings.warn(
2025-08-21 02:28:46,825 INFO worker.py:1918 -- Started a local Ray instance. View the dashboard at [1m[32mhttp://127.0.0.1:8265 [39m[22m
[36m(pid=443936)[0m /scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/transformers/utils/hub.py:111: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
[36m(pid=443936)[0m warnings.warn(
[36m(TaskRunner pid=443936)[0m TaskRunner hostname: c613-021.vista.tacc.utexas.edu, PID: 443936
[36m(TaskRunner pid=443936)[0m {'actor_rollout_ref': {'actor': {'checkpoint': {'load_contents': ['model',
[36m(TaskRunner pid=443936)[0m 'optimizer',
[36m(TaskRunner pid=443936)[0m 'extra'],
[36m(TaskRunner pid=443936)[0m 'save_contents': ['model',
[36m(TaskRunner pid=443936)[0m 'optimizer',
[36m(TaskRunner pid=443936)[0m 'extra']},
[36m(TaskRunner pid=443936)[0m 'clip_ratio': 0.2,
[36m(TaskRunner pid=443936)[0m 'clip_ratio_c': 3.0,
[36m(TaskRunner pid=443936)[0m 'clip_ratio_high': 0.2,
[36m(TaskRunner pid=443936)[0m 'clip_ratio_low': 0.2,
[36m(TaskRunner pid=443936)[0m 'entropy_checkpointing': False,
[36m(TaskRunner pid=443936)[0m 'entropy_coeff': 0,
[36m(TaskRunner pid=443936)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=443936)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=443936)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=443936)[0m 'offload_policy': False,
[36m(TaskRunner pid=443936)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=443936)[0m 'param_offload': False,
[36m(TaskRunner pid=443936)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=443936)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=443936)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=443936)[0m 'kl_loss_coef': 0.001,
[36m(TaskRunner pid=443936)[0m 'kl_loss_type': 'low_var_kl',
[36m(TaskRunner pid=443936)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=443936)[0m 'optim': {'lr': 1e-06,
[36m(TaskRunner pid=443936)[0m 'lr_warmup_steps': -1,
[36m(TaskRunner pid=443936)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=443936)[0m 'min_lr_ratio': 0.0,
[36m(TaskRunner pid=443936)[0m 'num_cycles': 0.5,
[36m(TaskRunner pid=443936)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=443936)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=443936)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=443936)[0m 'policy_loss': {'clip_cov_lb': 1.0,
[36m(TaskRunner pid=443936)[0m 'clip_cov_ratio': 0.0002,
[36m(TaskRunner pid=443936)[0m 'clip_cov_ub': 5.0,
[36m(TaskRunner pid=443936)[0m 'kl_cov_ratio': 0.0002,
[36m(TaskRunner pid=443936)[0m 'loss_mode': 'vanilla',
[36m(TaskRunner pid=443936)[0m 'ppo_kl_coef': 0.1},
[36m(TaskRunner pid=443936)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=443936)[0m 'ppo_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=443936)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=443936)[0m 'ppo_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=443936)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=443936)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=443936)[0m 'all_ranks': False,
[36m(TaskRunner pid=443936)[0m 'discrete': False,
[36m(TaskRunner pid=443936)[0m 'ranks': []},
[36m(TaskRunner pid=443936)[0m 'shuffle': False,
[36m(TaskRunner pid=443936)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=443936)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=443936)[0m 'use_dynamic_bsz': False,
[36m(TaskRunner pid=443936)[0m 'use_kl_loss': False,
[36m(TaskRunner pid=443936)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=443936)[0m 'hybrid_engine': True,
[36m(TaskRunner pid=443936)[0m 'model': {'custom_chat_template': None,
[36m(TaskRunner pid=443936)[0m 'enable_activation_offload': True,
[36m(TaskRunner pid=443936)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=443936)[0m 'exclude_modules': None,
[36m(TaskRunner pid=443936)[0m 'external_lib': None,
[36m(TaskRunner pid=443936)[0m 'fused_kernel_options': {'impl_backend': 'torch'},
[36m(TaskRunner pid=443936)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=443936)[0m 'lora_rank': 0,
[36m(TaskRunner pid=443936)[0m 'override_config': {},
[36m(TaskRunner pid=443936)[0m 'path': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=443936)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=443936)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=443936)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=443936)[0m 'use_liger': False,
[36m(TaskRunner pid=443936)[0m 'use_remove_padding': True,
[36m(TaskRunner pid=443936)[0m 'use_shm': False},
[36m(TaskRunner pid=443936)[0m 'ref': {'entropy_checkpointing': False,
[36m(TaskRunner pid=443936)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=443936)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=443936)[0m 'param_offload': False,
[36m(TaskRunner pid=443936)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=443936)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=443936)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=443936)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=443936)[0m 'log_prob_micro_batch_size_per_gpu': 32,
[36m(TaskRunner pid=443936)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=443936)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=443936)[0m 'all_ranks': False,
[36m(TaskRunner pid=443936)[0m 'discrete': False,
[36m(TaskRunner pid=443936)[0m 'ranks': []},
[36m(TaskRunner pid=443936)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=443936)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=443936)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=443936)[0m 'rollout': {'agent': {'custom_async_server': {'name': None,
[36m(TaskRunner pid=443936)[0m 'path': None},
[36m(TaskRunner pid=443936)[0m 'num_workers': 8},
[36m(TaskRunner pid=443936)[0m 'calculate_log_probs': False,
[36m(TaskRunner pid=443936)[0m 'disable_log_stats': True,
[36m(TaskRunner pid=443936)[0m 'do_sample': True,
[36m(TaskRunner pid=443936)[0m 'dtype': 'bfloat16',
[36m(TaskRunner pid=443936)[0m 'enable_chunked_prefill': True,
[36m(TaskRunner pid=443936)[0m 'enforce_eager': True,
[36m(TaskRunner pid=443936)[0m 'engine_kwargs': {'sglang': {'attention_backend': None},
[36m(TaskRunner pid=443936)[0m 'vllm': {'disable_mm_preprocessor_cache': False,
[36m(TaskRunner pid=443936)[0m 'swap_space': None}},
[36m(TaskRunner pid=443936)[0m 'free_cache_engine': True,
[36m(TaskRunner pid=443936)[0m 'gpu_memory_utilization': 0.8,
[36m(TaskRunner pid=443936)[0m 'ignore_eos': False,
[36m(TaskRunner pid=443936)[0m 'layered_summon': False,
[36m(TaskRunner pid=443936)[0m 'load_format': 'dummy_dtensor',
[36m(TaskRunner pid=443936)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=443936)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=443936)[0m 'log_prob_micro_batch_size_per_gpu': 32,
[36m(TaskRunner pid=443936)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=443936)[0m 'max_model_len': None,
[36m(TaskRunner pid=443936)[0m 'max_num_batched_tokens': 8192,
[36m(TaskRunner pid=443936)[0m 'max_num_seqs': 1024,
[36m(TaskRunner pid=443936)[0m 'mode': 'sync',
[36m(TaskRunner pid=443936)[0m 'multi_stage_wake_up': False,
[36m(TaskRunner pid=443936)[0m 'multi_turn': {'completion_callback': None,
[36m(TaskRunner pid=443936)[0m 'enable': False,
[36m(TaskRunner pid=443936)[0m 'format': 'hermes',
[36m(TaskRunner pid=443936)[0m 'interaction_config_path': None,
[36m(TaskRunner pid=443936)[0m 'max_assistant_turns': None,
[36m(TaskRunner pid=443936)[0m 'max_parallel_calls': 1,
[36m(TaskRunner pid=443936)[0m 'max_tool_response_length': 256,
[36m(TaskRunner pid=443936)[0m 'max_user_turns': None,
[36m(TaskRunner pid=443936)[0m 'tokenization_sanity_check_mode': 'strict',
[36m(TaskRunner pid=443936)[0m 'tool_config_path': None,
[36m(TaskRunner pid=443936)[0m 'tool_response_truncate_side': 'middle',
[36m(TaskRunner pid=443936)[0m 'use_inference_chat_template': False},
[36m(TaskRunner pid=443936)[0m 'n': 16,
[36m(TaskRunner pid=443936)[0m 'name': 'vllm',
[36m(TaskRunner pid=443936)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=443936)[0m 'all_ranks': False,
[36m(TaskRunner pid=443936)[0m 'discrete': False,
[36m(TaskRunner pid=443936)[0m 'ranks': []},
[36m(TaskRunner pid=443936)[0m 'prompt_length': 512,
[36m(TaskRunner pid=443936)[0m 'response_length': 4096,
[36m(TaskRunner pid=443936)[0m 'temperature': 1.0,
[36m(TaskRunner pid=443936)[0m 'tensor_model_parallel_size': 1,
[36m(TaskRunner pid=443936)[0m 'top_k': -1,
[36m(TaskRunner pid=443936)[0m 'top_p': 1,
[36m(TaskRunner pid=443936)[0m 'val_kwargs': {'do_sample': False,
[36m(TaskRunner pid=443936)[0m 'n': 1,
[36m(TaskRunner pid=443936)[0m 'temperature': 0,
[36m(TaskRunner pid=443936)[0m 'top_k': -1,
[36m(TaskRunner pid=443936)[0m 'top_p': 1.0}}},
[36m(TaskRunner pid=443936)[0m 'algorithm': {'_target_': 'verl.trainer.config.AlgoConfig',
[36m(TaskRunner pid=443936)[0m 'adv_estimator': 'grpo',
[36m(TaskRunner pid=443936)[0m 'gamma': 1.0,
[36m(TaskRunner pid=443936)[0m 'kl_ctrl': {'_target_': 'verl.trainer.config.KLControlConfig',
[36m(TaskRunner pid=443936)[0m 'horizon': 10000,
[36m(TaskRunner pid=443936)[0m 'kl_coef': 0.001,
[36m(TaskRunner pid=443936)[0m 'target_kl': 0.1,
[36m(TaskRunner pid=443936)[0m 'type': 'fixed'},
[36m(TaskRunner pid=443936)[0m 'kl_penalty': 'kl',
[36m(TaskRunner pid=443936)[0m 'lam': 1.0,
[36m(TaskRunner pid=443936)[0m 'norm_adv_by_std_in_grpo': True,
[36m(TaskRunner pid=443936)[0m 'pf_ppo': {'_target_': 'verl.trainer.config.PFPPOConfig',
[36m(TaskRunner pid=443936)[0m 'reweight_method': 'pow',
[36m(TaskRunner pid=443936)[0m 'weight_pow': 2.0},
[36m(TaskRunner pid=443936)[0m 'use_kl_in_reward': False,
[36m(TaskRunner pid=443936)[0m 'use_pf_ppo': False},
[36m(TaskRunner pid=443936)[0m 'critic': {'checkpoint': {'load_contents': ['model', 'optimizer', 'extra'],
[36m(TaskRunner pid=443936)[0m 'save_contents': ['model', 'optimizer', 'extra']},
[36m(TaskRunner pid=443936)[0m 'cliprange_value': 0.5,
[36m(TaskRunner pid=443936)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=443936)[0m 'forward_micro_batch_size': None,
[36m(TaskRunner pid=443936)[0m 'forward_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=443936)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=443936)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=443936)[0m 'model': {'enable_activation_offload': False,
[36m(TaskRunner pid=443936)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=443936)[0m 'external_lib': None,
[36m(TaskRunner pid=443936)[0m 'fsdp_config': {'forward_prefetch': False,
[36m(TaskRunner pid=443936)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=443936)[0m 'offload_policy': False,
[36m(TaskRunner pid=443936)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=443936)[0m 'param_offload': False,
[36m(TaskRunner pid=443936)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=443936)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=443936)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=443936)[0m 'lora_rank': 0,
[36m(TaskRunner pid=443936)[0m 'override_config': {},
[36m(TaskRunner pid=443936)[0m 'path': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=443936)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=443936)[0m 'tokenizer_path': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=443936)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=443936)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=443936)[0m 'use_shm': False},
[36m(TaskRunner pid=443936)[0m 'optim': {'lr': 1e-05,
[36m(TaskRunner pid=443936)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=443936)[0m 'min_lr_ratio': None,
[36m(TaskRunner pid=443936)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=443936)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=443936)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=443936)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=443936)[0m 'ppo_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=443936)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=443936)[0m 'ppo_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=443936)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=443936)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=443936)[0m 'all_ranks': False,
[36m(TaskRunner pid=443936)[0m 'discrete': False,
[36m(TaskRunner pid=443936)[0m 'ranks': []},
[36m(TaskRunner pid=443936)[0m 'rollout_n': 16,
[36m(TaskRunner pid=443936)[0m 'shuffle': False,
[36m(TaskRunner pid=443936)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=443936)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=443936)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=443936)[0m 'custom_reward_function': {'name': 'compute_score',
[36m(TaskRunner pid=443936)[0m 'path': '/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py',
[36m(TaskRunner pid=443936)[0m 'reward_kwargs': {'format_score_v2_weight': 0.0,
[36m(TaskRunner pid=443936)[0m 'format_score_weight': 0.0,
[36m(TaskRunner pid=443936)[0m 'reward_max': 10.0,
[36m(TaskRunner pid=443936)[0m 'reward_min': 0.0,
[36m(TaskRunner pid=443936)[0m 'transition_penalty_weight': 0.0}},
[36m(TaskRunner pid=443936)[0m 'data': {'custom_cls': {'name': None, 'path': None},
[36m(TaskRunner pid=443936)[0m 'dataloader_num_workers': 8,
[36m(TaskRunner pid=443936)[0m 'filter_overlong_prompts': False,
[36m(TaskRunner pid=443936)[0m 'filter_overlong_prompts_workers': 1,
[36m(TaskRunner pid=443936)[0m 'image_key': 'images',
[36m(TaskRunner pid=443936)[0m 'max_prompt_length': 512,
[36m(TaskRunner pid=443936)[0m 'max_response_length': 4096,
[36m(TaskRunner pid=443936)[0m 'prompt_key': 'prompt',
[36m(TaskRunner pid=443936)[0m 'return_full_prompt': False,
[36m(TaskRunner pid=443936)[0m 'return_multi_modal_inputs': True,
[36m(TaskRunner pid=443936)[0m 'return_raw_chat': False,
[36m(TaskRunner pid=443936)[0m 'return_raw_input_ids': False,
[36m(TaskRunner pid=443936)[0m 'reward_fn_key': 'data_source',
[36m(TaskRunner pid=443936)[0m 'sampler': {'class_name': None, 'class_path': None},
[36m(TaskRunner pid=443936)[0m 'shuffle': True,
[36m(TaskRunner pid=443936)[0m 'tokenizer': None,
[36m(TaskRunner pid=443936)[0m 'train_batch_size': 256,
[36m(TaskRunner pid=443936)[0m 'train_files': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/train.parquet',
[36m(TaskRunner pid=443936)[0m 'truncation': 'error',
[36m(TaskRunner pid=443936)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=443936)[0m 'use_shm': False,
[36m(TaskRunner pid=443936)[0m 'val_batch_size': None,
[36m(TaskRunner pid=443936)[0m 'val_files': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/test.parquet',
[36m(TaskRunner pid=443936)[0m 'validation_shuffle': False,
[36m(TaskRunner pid=443936)[0m 'video_key': 'videos'},
[36m(TaskRunner pid=443936)[0m 'ray_init': {'num_cpus': None, 'timeline_json_file': None},
[36m(TaskRunner pid=443936)[0m 'reward_model': {'enable': False,
[36m(TaskRunner pid=443936)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=443936)[0m 'launch_reward_fn_async': False,
[36m(TaskRunner pid=443936)[0m 'max_length': None,
[36m(TaskRunner pid=443936)[0m 'micro_batch_size': None,
[36m(TaskRunner pid=443936)[0m 'micro_batch_size_per_gpu': None,
[36m(TaskRunner pid=443936)[0m 'model': {'external_lib': None,
[36m(TaskRunner pid=443936)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=443936)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=443936)[0m 'param_offload': False,
[36m(TaskRunner pid=443936)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=443936)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=443936)[0m 'input_tokenizer': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=443936)[0m 'path': '~/models/FsfairX-LLaMA3-RM-v0.1',
[36m(TaskRunner pid=443936)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=443936)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=443936)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=443936)[0m 'use_shm': False},
[36m(TaskRunner pid=443936)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=443936)[0m 'all_ranks': False,
[36m(TaskRunner pid=443936)[0m 'discrete': False,
[36m(TaskRunner pid=443936)[0m 'ranks': []},
[36m(TaskRunner pid=443936)[0m 'reward_manager': 'naive',
[36m(TaskRunner pid=443936)[0m 'sandbox_fusion': {'max_concurrent': 64,
[36m(TaskRunner pid=443936)[0m 'memory_limit_mb': 1024,
[36m(TaskRunner pid=443936)[0m 'url': None},
[36m(TaskRunner pid=443936)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=443936)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=443936)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=443936)[0m 'trainer': {'balance_batch': True,
[36m(TaskRunner pid=443936)[0m 'controller_nsight_options': {'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=443936)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=443936)[0m 'trace': 'cuda,nvtx,cublas,ucx'},
[36m(TaskRunner pid=443936)[0m 'critic_warmup': 0,
[36m(TaskRunner pid=443936)[0m 'default_hdfs_dir': None,
[36m(TaskRunner pid=443936)[0m 'default_local_dir': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints',
[36m(TaskRunner pid=443936)[0m 'del_local_ckpt_after_load': False,
[36m(TaskRunner pid=443936)[0m 'device': 'cuda',
[36m(TaskRunner pid=443936)[0m 'esi_redundant_time': 0,
[36m(TaskRunner pid=443936)[0m 'experiment_name': '0818_rl_alltask_tacc__test_rolloutn16_1gpu_rl',
[36m(TaskRunner pid=443936)[0m 'log_val_generations': 0,
[36m(TaskRunner pid=443936)[0m 'logger': ['console', 'wandb'],
[36m(TaskRunner pid=443936)[0m 'max_actor_ckpt_to_keep': None,
[36m(TaskRunner pid=443936)[0m 'max_critic_ckpt_to_keep': None,
[36m(TaskRunner pid=443936)[0m 'n_gpus_per_node': 1,
[36m(TaskRunner pid=443936)[0m 'nnodes': 1,
[36m(TaskRunner pid=443936)[0m 'profile_steps': None,
[36m(TaskRunner pid=443936)[0m 'project_name': 'jackrl',
[36m(TaskRunner pid=443936)[0m 'ray_wait_register_center_timeout': 300,
[36m(TaskRunner pid=443936)[0m 'resume_from_path': None,
[36m(TaskRunner pid=443936)[0m 'resume_mode': 'disable',
[36m(TaskRunner pid=443936)[0m 'rollout_data_dir': None,
[36m(TaskRunner pid=443936)[0m 'save_freq': 15,
[36m(TaskRunner pid=443936)[0m 'test_freq': 5,
[36m(TaskRunner pid=443936)[0m 'total_epochs': 8,
[36m(TaskRunner pid=443936)[0m 'total_training_steps': None,
[36m(TaskRunner pid=443936)[0m 'val_before_train': False,
[36m(TaskRunner pid=443936)[0m 'val_only': False,
[36m(TaskRunner pid=443936)[0m 'validation_data_dir': None,
[36m(TaskRunner pid=443936)[0m 'worker_nsight_options': {'capture-range': 'cudaProfilerApi',
[36m(TaskRunner pid=443936)[0m 'capture-range-end': None,
[36m(TaskRunner pid=443936)[0m 'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=443936)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=443936)[0m 'kill': 'none',
[36m(TaskRunner pid=443936)[0m 'trace': 'cuda,nvtx,cublas,ucx'}}}
[36m(TaskRunner pid=443936)[0m Registered source: longmult
[36m(TaskRunner pid=443936)[0m Registered source: countdown
[36m(TaskRunner pid=443936)[0m Registered source: gsm8k
[36m(TaskRunner pid=443936)[0m Registered source: arc
[36m(TaskRunner pid=443936)[0m Registered source: arc_challenge
[36m(TaskRunner pid=443936)[0m Registered source: arc_easy
[36m(TaskRunner pid=443936)[0m Registered source: piqa
[36m(TaskRunner pid=443936)[0m Registered source: mmlu
[36m(TaskRunner pid=443936)[0m Registered source: mmlu_pro
[36m(TaskRunner pid=443936)[0m Registered source: csqa
[36m(TaskRunner pid=443936)[0m Registered source: social_iqa
[36m(TaskRunner pid=443936)[0m Registered source: strategy_qa
[36m(TaskRunner pid=443936)[0m Registered source: winogrande
[36m(TaskRunner pid=443936)[0m Registered source: bbh
[36m(TaskRunner pid=443936)[0m using customized reward function 'compute_score' from '/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=443936)[0m using customized reward function 'compute_score' from '/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=443936)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=443936)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=443936)[0m
Generating train split: 1000 examples [00:00, 8675.15 examples/s]
[36m(TaskRunner pid=443936)[0m
Generating train split: 4000 examples [00:00, 8472.03 examples/s]
[36m(TaskRunner pid=443936)[0m dataset len: 4000
[36m(TaskRunner pid=443936)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=443936)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=443936)[0m
Generating train split: 1000 examples [00:00, 9600.65 examples/s]
[36m(TaskRunner pid=443936)[0m dataset len: 1000
[36m(TaskRunner pid=443936)[0m Using critic: False
[36m(TaskRunner pid=443936)[0m [validate_config] All configuration checks passed successfully!
[36m(TaskRunner pid=443936)[0m Size of train dataloader: 15, Size of val dataloader: 1
[36m(TaskRunner pid=443936)[0m Total training steps: 120
[36m(TaskRunner pid=443936)[0m {'de5032a7e1f6795907086a69f9e95758f8c5f44cf01f268e6a9c8a09': {'CPU': 71.0,
[36m(TaskRunner pid=443936)[0m 'GPU': 1.0,
[36m(TaskRunner pid=443936)[0m 'accelerator_type:GH200': 1.0,
[36m(TaskRunner pid=443936)[0m 'memory': 158863694233.0,
[36m(TaskRunner pid=443936)[0m 'node:129.114.17.164': 1.0,
[36m(TaskRunner pid=443936)[0m 'node:__internal_head__': 1.0,
[36m(TaskRunner pid=443936)[0m 'object_store_memory': 60055971430.0}}
[36m(TaskRunner pid=443936)[0m ('Resource pool to cls: {<verl.single_controller.ray.base.RayResourcePool '
[36m(TaskRunner pid=443936)[0m "object at 0x400f314bad50>: {'actor_rollout': "
[36m(TaskRunner pid=443936)[0m '<verl.single_controller.ray.base.RayClassWithInitArgs object at '
[36m(TaskRunner pid=443936)[0m '0x400f1fa05dd0>}}')
[36m(TaskRunner pid=443936)[0m colocated worker base class <class 'verl.single_controller.base.worker.Worker'>
[36m(TaskRunner pid=443936)[0m DeprecationWarning: `ray.state.available_resources_per_node` is a private attribute and access will be removed in a future Ray version.
[36m(TaskRunner pid=443936)[0m WARNING:2025-08-21 02:29:14,760:Waiting for register center actor xIvSFo_register_center to be ready. Elapsed time: 0 seconds out of 300 seconds.
[36m(pid=445525)[0m /scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/transformers/utils/hub.py:111: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
[36m(pid=445525)[0m warnings.warn(
[36m(WorkerDict pid=445525)[0m Model config after override: Qwen2Config {
[36m(WorkerDict pid=445525)[0m "architectures": [
[36m(WorkerDict pid=445525)[0m "Qwen2ForCausalLM"
[36m(WorkerDict pid=445525)[0m ],
[36m(WorkerDict pid=445525)[0m "attention_dropout": 0.0,
[36m(WorkerDict pid=445525)[0m "eos_token_id": 151645,
[36m(WorkerDict pid=445525)[0m "hidden_act": "silu",
[36m(WorkerDict pid=445525)[0m "hidden_size": 1536,
[36m(WorkerDict pid=445525)[0m "initializer_range": 0.02,
[36m(WorkerDict pid=445525)[0m "intermediate_size": 8960,
[36m(WorkerDict pid=445525)[0m "max_position_embeddings": 32768,
[36m(WorkerDict pid=445525)[0m "max_window_layers": 21,
[36m(WorkerDict pid=445525)[0m "model_type": "qwen2",
[36m(WorkerDict pid=445525)[0m "num_attention_heads": 12,
[36m(WorkerDict pid=445525)[0m "num_hidden_layers": 28,
[36m(WorkerDict pid=445525)[0m "num_key_value_heads": 2,
[36m(WorkerDict pid=445525)[0m "pad_token_id": 151643,
[36m(WorkerDict pid=445525)[0m "rms_norm_eps": 1e-06,
[36m(WorkerDict pid=445525)[0m "rope_scaling": null,
[36m(WorkerDict pid=445525)[0m "rope_theta": 1000000.0,
[36m(WorkerDict pid=445525)[0m "sliding_window": 32768,
[36m(WorkerDict pid=445525)[0m "tie_word_embeddings": true,
[36m(WorkerDict pid=445525)[0m "torch_dtype": "bfloat16",
[36m(WorkerDict pid=445525)[0m "transformers_version": "4.52.3",
[36m(WorkerDict pid=445525)[0m "use_cache": true,
[36m(WorkerDict pid=445525)[0m "use_sliding_window": false,
[36m(WorkerDict pid=445525)[0m "vocab_size": 151936
[36m(WorkerDict pid=445525)[0m }
[36m(WorkerDict pid=445525)[0m
[36m(WorkerDict pid=445525)[0m Flash Attention 2.0 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen2ForCausalLM is torch.float32. You should run training or inference using Automatic Mixed-Precision via the `with torch.autocast(device_type='torch_device'):` decorator, or load the model with the `torch_dtype` argument. Example: `model = AutoModel.from_pretrained("openai/whisper-tiny", attn_implementation="flash_attention_2", torch_dtype=torch.float16)`
[36m(WorkerDict pid=445525)[0m You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
[36m(WorkerDict pid=445525)[0m Monkey patch _flash_attention_forward in transformers.integrations.flash_attention
[36m(WorkerDict pid=445525)[0m Skipping monkey patch for Qwen2ForCausalLM as use_fused_kernels is False or fused_kernels_backend is torch
[36m(WorkerDict pid=445525)[0m Qwen2ForCausalLM contains 1.54B parameters
[36m(WorkerDict pid=445525)[0m wrap_policy: functools.partial(<function _or_policy at 0x400ef5ec6e80>, policies=[functools.partial(<function transformer_auto_wrap_policy at 0x400ef5ec6d40>, transformer_layer_cls={<class 'transformers.models.qwen2.modeling_qwen2.Qwen2DecoderLayer'>})])
[36m(WorkerDict pid=445525)[0m
[36m(WorkerDict pid=445525)[0m c613-021:445525:445525 [0] bootstrap.cc:48 NCCL WARN Bootstrap : no socket interface found
Error executing job with overrides: ['trainer.total_epochs=8', 'actor_rollout_ref.actor.optim.lr=1e-06', 'trainer.save_freq=15', 'trainer.test_freq=5', 'trainer.val_before_train=False', 'algorithm.adv_estimator=grpo', 'actor_rollout_ref.rollout.n=16', 'data.train_batch_size=256', 'actor_rollout_ref.actor.ppo_mini_batch_size=32', 'actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=16', 'actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32', 'actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32', 'custom_reward_function.reward_kwargs.format_score_weight=0.0', 'custom_reward_function.reward_kwargs.format_score_v2_weight=0.0', 'custom_reward_function.reward_kwargs.transition_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.reward_min=0.0', 'custom_reward_function.reward_kwargs.reward_max=10.0', 'actor_rollout_ref.model.enable_gradient_checkpointing=True', 'actor_rollout_ref.model.enable_activation_offload=True', 'actor_rollout_ref.rollout.gpu_memory_utilization=0.8', 'actor_rollout_ref.model.use_remove_padding=True', 'actor_rollout_ref.actor.strategy=fsdp2', 'actor_rollout_ref.actor.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.ref.fsdp_config.forward_prefetch=True', 'reward_model.model.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.rollout.tensor_model_parallel_size=1', 'data.max_prompt_length=512', 'data.max_response_length=4096', 'actor_rollout_ref.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft', 'actor_rollout_ref.rollout.dtype=bfloat16', 'critic.optim.lr=1e-05', 'critic.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft', 'critic.ppo_micro_batch_size_per_gpu=1', 'algorithm.kl_ctrl.kl_coef=0.001', 'trainer.logger=[console,wandb]', 'trainer.project_name=jackrl', 'trainer.experiment_name=0818_rl_alltask_tacc__test_rolloutn16_1gpu_rl', 'trainer.resume_mode=disable', 'data.train_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/train.parquet', 'data.val_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/test.parquet', 'custom_reward_function.path=/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py', 'trainer.default_local_dir=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints', 'actor_rollout_ref.model.trust_remote_code=True', 'critic.model.trust_remote_code=True', 'trainer.nnodes=1', 'trainer.n_gpus_per_node=1']
Traceback (most recent call last):
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 39, in main
run_ppo(config)
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 69, in run_ppo
ray.get(runner.run.remote(config))
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/worker.py", line 2858, in get
values, debugger_breakpoint = worker.get_objects(object_refs, timeout=timeout)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/worker.py", line 958, in get_objects
raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(DistBackendError): [36mray::TaskRunner.run()[39m (pid=443936, ip=129.114.17.164, actor_id=ac2c5aa8565ad98c28f6860901000000, repr=<main_ppo.TaskRunner object at 0x400f1f782b10>)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 232, in run
trainer.init_workers()
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/ppo/ray_trainer.py", line 931, in init_workers
self.actor_rollout_wg.init_model()
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 51, in __call__
output = ray.get(output)
^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ray.exceptions.RayTaskError(DistBackendError): [36mray::WorkerDict.actor_rollout_init_model()[39m (pid=445525, ip=129.114.17.164, actor_id=6146d09415258dd8477cea5701000000, repr=<verl.single_controller.ray.base.WorkerDict object at 0x401142604c90>)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 708, in func
return getattr(self.worker_dict[key], name)(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/decorator.py", line 549, in inner
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/workers/fsdp_workers.py", line 594, in init_model
) = self._build_model_optimizer(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/workers/fsdp_workers.py", line 408, in _build_model_optimizer
fsdp2_load_full_state_dict(actor_module, full_state, fsdp_mesh, cpu_offload)
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/utils/fsdp_utils.py", line 466, in fsdp2_load_full_state_dict
set_model_state_dict(model, full_state, options=options)
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/checkpoint/state_dict.py", line 1207, in set_model_state_dict
return _load_model_state_dict(model, model_state_dict, info)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/checkpoint/state_dict.py", line 569, in _load_model_state_dict
_broadcast_state_dict(
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/_state_dict_utils.py", line 616, in _broadcast_state_dict
_broadcast_tensors(ret, local_state_dict, keys, device, pg)
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/_state_dict_utils.py", line 535, in _broadcast_tensors
dist.broadcast(tensors[0], src=0, group=pg)
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/c10d_logger.py", line 81, in wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/distributed_c10d.py", line 2726, in broadcast
work = group.broadcast([tensor], opts)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.distributed.DistBackendError: NCCL error in: /pytorch/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:2588, internal error - please report this issue to the NCCL developers, NCCL version 2.21.5
ncclInternalError: Internal check failed.
Last error:
Bootstrap : no socket interface found
Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.
[INFO] Extracting model from VeRL checkpoint at /scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints
[ERROR] No global_step directories found
EXTRACT OUT: False
[ERROR] Stage error: RuntimeError: Model extraction failed
|
/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/huggingface_hub/file_download.py:980: UserWarning: `local_dir_use_symlinks` parameter is deprecated and will be ignored. The process to download files to a local folder has been updated and do not rely on symlinks anymore. You only need to pass a destination folder as`local_dir`.
For more details, check out https://huggingface.co/docs/huggingface_hub/main/en/guides/download#download-files-to-local-folder.
warnings.warn(
Fetching 12 files: 0%| | 0/12 [00:00<?, ?it/s]
Fetching 12 files: 100%|███████████████████████████████████████████████████████████████████████| 12/12 [00:00<00:00, 586.19it/s]
Fetching 12 files: 0%| | 0/12 [00:00<?, ?it/s]
Fetching 12 files: 100%|███████████████████████████████████████████████████████████████████████| 12/12 [00:00<00:00, 373.91it/s]
|
0818_rl_alltask_tacc__test_rolloutn16_1gpu
| 63.320029 | true |
2025-08-21T02:30:51.883623
|
2025-08-21T02:31:34.631777
|
verl_rl
| 1 |
INFO
|
Complete log capture for stage: verl_rl
|
[INFO] Starting stage: VeRL RL training - rl
[INFO] Data preparation succeeded
[DEBUG] Running verl command:
python -m verl.trainer.main_ppo trainer.total_epochs=8 actor_rollout_ref.actor.optim.lr=1e-06 trainer.save_freq=15 trainer.test_freq=5 trainer.val_before_train=False algorithm.adv_estimator=grpo actor_rollout_ref.rollout.n=16 data.train_batch_size=256 actor_rollout_ref.actor.ppo_mini_batch_size=32 actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8 actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16 actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16 custom_reward_function.reward_kwargs.format_score_weight=0.0 custom_reward_function.reward_kwargs.format_score_v2_weight=0.0 custom_reward_function.reward_kwargs.transition_penalty_weight=0.0 custom_reward_function.reward_kwargs.reward_min=0.0 custom_reward_function.reward_kwargs.reward_max=10.0 actor_rollout_ref.model.enable_gradient_checkpointing=True actor_rollout_ref.model.enable_activation_offload=True actor_rollout_ref.rollout.gpu_memory_utilization=0.8 actor_rollout_ref.model.use_remove_padding=True actor_rollout_ref.actor.strategy=fsdp2 actor_rollout_ref.actor.fsdp_config.forward_prefetch=True actor_rollout_ref.ref.fsdp_config.forward_prefetch=True reward_model.model.fsdp_config.forward_prefetch=True hydra.run.dir=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/hydra hydra.output_subdir=null hydra.job.chdir=False actor_rollout_ref.rollout.tensor_model_parallel_size=1 data.max_prompt_length=512 data.max_response_length=4096 actor_rollout_ref.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft actor_rollout_ref.rollout.dtype=bfloat16 critic.optim.lr=1e-05 critic.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft critic.ppo_micro_batch_size_per_gpu=1 algorithm.kl_ctrl.kl_coef=0.001 trainer.logger=[console,wandb] trainer.project_name=jackrl trainer.experiment_name=0818_rl_alltask_tacc__test_rolloutn16_1gpu_rl trainer.resume_mode=disable data.train_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/train.parquet data.val_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/test.parquet custom_reward_function.path=/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py trainer.default_local_dir=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints actor_rollout_ref.model.trust_remote_code=True critic.model.trust_remote_code=True trainer.nnodes=1 trainer.n_gpus_per_node=1
/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/transformers/utils/hub.py:111: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
warnings.warn(
2025-08-21 02:31:10,691 INFO worker.py:1918 -- Started a local Ray instance. View the dashboard at [1m[32mhttp://127.0.0.1:8265 [39m[22m
[36m(pid=450885)[0m /scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/transformers/utils/hub.py:111: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
[36m(pid=450885)[0m warnings.warn(
[36m(TaskRunner pid=450885)[0m TaskRunner hostname: c613-021.vista.tacc.utexas.edu, PID: 450885
[36m(TaskRunner pid=450885)[0m {'actor_rollout_ref': {'actor': {'checkpoint': {'load_contents': ['model',
[36m(TaskRunner pid=450885)[0m 'optimizer',
[36m(TaskRunner pid=450885)[0m 'extra'],
[36m(TaskRunner pid=450885)[0m 'save_contents': ['model',
[36m(TaskRunner pid=450885)[0m 'optimizer',
[36m(TaskRunner pid=450885)[0m 'extra']},
[36m(TaskRunner pid=450885)[0m 'clip_ratio': 0.2,
[36m(TaskRunner pid=450885)[0m 'clip_ratio_c': 3.0,
[36m(TaskRunner pid=450885)[0m 'clip_ratio_high': 0.2,
[36m(TaskRunner pid=450885)[0m 'clip_ratio_low': 0.2,
[36m(TaskRunner pid=450885)[0m 'entropy_checkpointing': False,
[36m(TaskRunner pid=450885)[0m 'entropy_coeff': 0,
[36m(TaskRunner pid=450885)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=450885)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=450885)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=450885)[0m 'offload_policy': False,
[36m(TaskRunner pid=450885)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=450885)[0m 'param_offload': False,
[36m(TaskRunner pid=450885)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=450885)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=450885)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=450885)[0m 'kl_loss_coef': 0.001,
[36m(TaskRunner pid=450885)[0m 'kl_loss_type': 'low_var_kl',
[36m(TaskRunner pid=450885)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=450885)[0m 'optim': {'lr': 1e-06,
[36m(TaskRunner pid=450885)[0m 'lr_warmup_steps': -1,
[36m(TaskRunner pid=450885)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=450885)[0m 'min_lr_ratio': 0.0,
[36m(TaskRunner pid=450885)[0m 'num_cycles': 0.5,
[36m(TaskRunner pid=450885)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=450885)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=450885)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=450885)[0m 'policy_loss': {'clip_cov_lb': 1.0,
[36m(TaskRunner pid=450885)[0m 'clip_cov_ratio': 0.0002,
[36m(TaskRunner pid=450885)[0m 'clip_cov_ub': 5.0,
[36m(TaskRunner pid=450885)[0m 'kl_cov_ratio': 0.0002,
[36m(TaskRunner pid=450885)[0m 'loss_mode': 'vanilla',
[36m(TaskRunner pid=450885)[0m 'ppo_kl_coef': 0.1},
[36m(TaskRunner pid=450885)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=450885)[0m 'ppo_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=450885)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=450885)[0m 'ppo_micro_batch_size_per_gpu': 8,
[36m(TaskRunner pid=450885)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=450885)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=450885)[0m 'all_ranks': False,
[36m(TaskRunner pid=450885)[0m 'discrete': False,
[36m(TaskRunner pid=450885)[0m 'ranks': []},
[36m(TaskRunner pid=450885)[0m 'shuffle': False,
[36m(TaskRunner pid=450885)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=450885)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=450885)[0m 'use_dynamic_bsz': False,
[36m(TaskRunner pid=450885)[0m 'use_kl_loss': False,
[36m(TaskRunner pid=450885)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=450885)[0m 'hybrid_engine': True,
[36m(TaskRunner pid=450885)[0m 'model': {'custom_chat_template': None,
[36m(TaskRunner pid=450885)[0m 'enable_activation_offload': True,
[36m(TaskRunner pid=450885)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=450885)[0m 'exclude_modules': None,
[36m(TaskRunner pid=450885)[0m 'external_lib': None,
[36m(TaskRunner pid=450885)[0m 'fused_kernel_options': {'impl_backend': 'torch'},
[36m(TaskRunner pid=450885)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=450885)[0m 'lora_rank': 0,
[36m(TaskRunner pid=450885)[0m 'override_config': {},
[36m(TaskRunner pid=450885)[0m 'path': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=450885)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=450885)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=450885)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=450885)[0m 'use_liger': False,
[36m(TaskRunner pid=450885)[0m 'use_remove_padding': True,
[36m(TaskRunner pid=450885)[0m 'use_shm': False},
[36m(TaskRunner pid=450885)[0m 'ref': {'entropy_checkpointing': False,
[36m(TaskRunner pid=450885)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=450885)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=450885)[0m 'param_offload': False,
[36m(TaskRunner pid=450885)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=450885)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=450885)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=450885)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=450885)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=450885)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=450885)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=450885)[0m 'all_ranks': False,
[36m(TaskRunner pid=450885)[0m 'discrete': False,
[36m(TaskRunner pid=450885)[0m 'ranks': []},
[36m(TaskRunner pid=450885)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=450885)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=450885)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=450885)[0m 'rollout': {'agent': {'custom_async_server': {'name': None,
[36m(TaskRunner pid=450885)[0m 'path': None},
[36m(TaskRunner pid=450885)[0m 'num_workers': 8},
[36m(TaskRunner pid=450885)[0m 'calculate_log_probs': False,
[36m(TaskRunner pid=450885)[0m 'disable_log_stats': True,
[36m(TaskRunner pid=450885)[0m 'do_sample': True,
[36m(TaskRunner pid=450885)[0m 'dtype': 'bfloat16',
[36m(TaskRunner pid=450885)[0m 'enable_chunked_prefill': True,
[36m(TaskRunner pid=450885)[0m 'enforce_eager': True,
[36m(TaskRunner pid=450885)[0m 'engine_kwargs': {'sglang': {'attention_backend': None},
[36m(TaskRunner pid=450885)[0m 'vllm': {'disable_mm_preprocessor_cache': False,
[36m(TaskRunner pid=450885)[0m 'swap_space': None}},
[36m(TaskRunner pid=450885)[0m 'free_cache_engine': True,
[36m(TaskRunner pid=450885)[0m 'gpu_memory_utilization': 0.8,
[36m(TaskRunner pid=450885)[0m 'ignore_eos': False,
[36m(TaskRunner pid=450885)[0m 'layered_summon': False,
[36m(TaskRunner pid=450885)[0m 'load_format': 'dummy_dtensor',
[36m(TaskRunner pid=450885)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=450885)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=450885)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=450885)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=450885)[0m 'max_model_len': None,
[36m(TaskRunner pid=450885)[0m 'max_num_batched_tokens': 8192,
[36m(TaskRunner pid=450885)[0m 'max_num_seqs': 1024,
[36m(TaskRunner pid=450885)[0m 'mode': 'sync',
[36m(TaskRunner pid=450885)[0m 'multi_stage_wake_up': False,
[36m(TaskRunner pid=450885)[0m 'multi_turn': {'completion_callback': None,
[36m(TaskRunner pid=450885)[0m 'enable': False,
[36m(TaskRunner pid=450885)[0m 'format': 'hermes',
[36m(TaskRunner pid=450885)[0m 'interaction_config_path': None,
[36m(TaskRunner pid=450885)[0m 'max_assistant_turns': None,
[36m(TaskRunner pid=450885)[0m 'max_parallel_calls': 1,
[36m(TaskRunner pid=450885)[0m 'max_tool_response_length': 256,
[36m(TaskRunner pid=450885)[0m 'max_user_turns': None,
[36m(TaskRunner pid=450885)[0m 'tokenization_sanity_check_mode': 'strict',
[36m(TaskRunner pid=450885)[0m 'tool_config_path': None,
[36m(TaskRunner pid=450885)[0m 'tool_response_truncate_side': 'middle',
[36m(TaskRunner pid=450885)[0m 'use_inference_chat_template': False},
[36m(TaskRunner pid=450885)[0m 'n': 16,
[36m(TaskRunner pid=450885)[0m 'name': 'vllm',
[36m(TaskRunner pid=450885)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=450885)[0m 'all_ranks': False,
[36m(TaskRunner pid=450885)[0m 'discrete': False,
[36m(TaskRunner pid=450885)[0m 'ranks': []},
[36m(TaskRunner pid=450885)[0m 'prompt_length': 512,
[36m(TaskRunner pid=450885)[0m 'response_length': 4096,
[36m(TaskRunner pid=450885)[0m 'temperature': 1.0,
[36m(TaskRunner pid=450885)[0m 'tensor_model_parallel_size': 1,
[36m(TaskRunner pid=450885)[0m 'top_k': -1,
[36m(TaskRunner pid=450885)[0m 'top_p': 1,
[36m(TaskRunner pid=450885)[0m 'val_kwargs': {'do_sample': False,
[36m(TaskRunner pid=450885)[0m 'n': 1,
[36m(TaskRunner pid=450885)[0m 'temperature': 0,
[36m(TaskRunner pid=450885)[0m 'top_k': -1,
[36m(TaskRunner pid=450885)[0m 'top_p': 1.0}}},
[36m(TaskRunner pid=450885)[0m 'algorithm': {'_target_': 'verl.trainer.config.AlgoConfig',
[36m(TaskRunner pid=450885)[0m 'adv_estimator': 'grpo',
[36m(TaskRunner pid=450885)[0m 'gamma': 1.0,
[36m(TaskRunner pid=450885)[0m 'kl_ctrl': {'_target_': 'verl.trainer.config.KLControlConfig',
[36m(TaskRunner pid=450885)[0m 'horizon': 10000,
[36m(TaskRunner pid=450885)[0m 'kl_coef': 0.001,
[36m(TaskRunner pid=450885)[0m 'target_kl': 0.1,
[36m(TaskRunner pid=450885)[0m 'type': 'fixed'},
[36m(TaskRunner pid=450885)[0m 'kl_penalty': 'kl',
[36m(TaskRunner pid=450885)[0m 'lam': 1.0,
[36m(TaskRunner pid=450885)[0m 'norm_adv_by_std_in_grpo': True,
[36m(TaskRunner pid=450885)[0m 'pf_ppo': {'_target_': 'verl.trainer.config.PFPPOConfig',
[36m(TaskRunner pid=450885)[0m 'reweight_method': 'pow',
[36m(TaskRunner pid=450885)[0m 'weight_pow': 2.0},
[36m(TaskRunner pid=450885)[0m 'use_kl_in_reward': False,
[36m(TaskRunner pid=450885)[0m 'use_pf_ppo': False},
[36m(TaskRunner pid=450885)[0m 'critic': {'checkpoint': {'load_contents': ['model', 'optimizer', 'extra'],
[36m(TaskRunner pid=450885)[0m 'save_contents': ['model', 'optimizer', 'extra']},
[36m(TaskRunner pid=450885)[0m 'cliprange_value': 0.5,
[36m(TaskRunner pid=450885)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=450885)[0m 'forward_micro_batch_size': None,
[36m(TaskRunner pid=450885)[0m 'forward_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=450885)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=450885)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=450885)[0m 'model': {'enable_activation_offload': False,
[36m(TaskRunner pid=450885)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=450885)[0m 'external_lib': None,
[36m(TaskRunner pid=450885)[0m 'fsdp_config': {'forward_prefetch': False,
[36m(TaskRunner pid=450885)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=450885)[0m 'offload_policy': False,
[36m(TaskRunner pid=450885)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=450885)[0m 'param_offload': False,
[36m(TaskRunner pid=450885)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=450885)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=450885)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=450885)[0m 'lora_rank': 0,
[36m(TaskRunner pid=450885)[0m 'override_config': {},
[36m(TaskRunner pid=450885)[0m 'path': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=450885)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=450885)[0m 'tokenizer_path': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=450885)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=450885)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=450885)[0m 'use_shm': False},
[36m(TaskRunner pid=450885)[0m 'optim': {'lr': 1e-05,
[36m(TaskRunner pid=450885)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=450885)[0m 'min_lr_ratio': None,
[36m(TaskRunner pid=450885)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=450885)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=450885)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=450885)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=450885)[0m 'ppo_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=450885)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=450885)[0m 'ppo_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=450885)[0m 'ppo_mini_batch_size': 32,
[36m(TaskRunner pid=450885)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=450885)[0m 'all_ranks': False,
[36m(TaskRunner pid=450885)[0m 'discrete': False,
[36m(TaskRunner pid=450885)[0m 'ranks': []},
[36m(TaskRunner pid=450885)[0m 'rollout_n': 16,
[36m(TaskRunner pid=450885)[0m 'shuffle': False,
[36m(TaskRunner pid=450885)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=450885)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=450885)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=450885)[0m 'custom_reward_function': {'name': 'compute_score',
[36m(TaskRunner pid=450885)[0m 'path': '/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py',
[36m(TaskRunner pid=450885)[0m 'reward_kwargs': {'format_score_v2_weight': 0.0,
[36m(TaskRunner pid=450885)[0m 'format_score_weight': 0.0,
[36m(TaskRunner pid=450885)[0m 'reward_max': 10.0,
[36m(TaskRunner pid=450885)[0m 'reward_min': 0.0,
[36m(TaskRunner pid=450885)[0m 'transition_penalty_weight': 0.0}},
[36m(TaskRunner pid=450885)[0m 'data': {'custom_cls': {'name': None, 'path': None},
[36m(TaskRunner pid=450885)[0m 'dataloader_num_workers': 8,
[36m(TaskRunner pid=450885)[0m 'filter_overlong_prompts': False,
[36m(TaskRunner pid=450885)[0m 'filter_overlong_prompts_workers': 1,
[36m(TaskRunner pid=450885)[0m 'image_key': 'images',
[36m(TaskRunner pid=450885)[0m 'max_prompt_length': 512,
[36m(TaskRunner pid=450885)[0m 'max_response_length': 4096,
[36m(TaskRunner pid=450885)[0m 'prompt_key': 'prompt',
[36m(TaskRunner pid=450885)[0m 'return_full_prompt': False,
[36m(TaskRunner pid=450885)[0m 'return_multi_modal_inputs': True,
[36m(TaskRunner pid=450885)[0m 'return_raw_chat': False,
[36m(TaskRunner pid=450885)[0m 'return_raw_input_ids': False,
[36m(TaskRunner pid=450885)[0m 'reward_fn_key': 'data_source',
[36m(TaskRunner pid=450885)[0m 'sampler': {'class_name': None, 'class_path': None},
[36m(TaskRunner pid=450885)[0m 'shuffle': True,
[36m(TaskRunner pid=450885)[0m 'tokenizer': None,
[36m(TaskRunner pid=450885)[0m 'train_batch_size': 256,
[36m(TaskRunner pid=450885)[0m 'train_files': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/train.parquet',
[36m(TaskRunner pid=450885)[0m 'truncation': 'error',
[36m(TaskRunner pid=450885)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=450885)[0m 'use_shm': False,
[36m(TaskRunner pid=450885)[0m 'val_batch_size': None,
[36m(TaskRunner pid=450885)[0m 'val_files': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/test.parquet',
[36m(TaskRunner pid=450885)[0m 'validation_shuffle': False,
[36m(TaskRunner pid=450885)[0m 'video_key': 'videos'},
[36m(TaskRunner pid=450885)[0m 'ray_init': {'num_cpus': None, 'timeline_json_file': None},
[36m(TaskRunner pid=450885)[0m 'reward_model': {'enable': False,
[36m(TaskRunner pid=450885)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=450885)[0m 'launch_reward_fn_async': False,
[36m(TaskRunner pid=450885)[0m 'max_length': None,
[36m(TaskRunner pid=450885)[0m 'micro_batch_size': None,
[36m(TaskRunner pid=450885)[0m 'micro_batch_size_per_gpu': None,
[36m(TaskRunner pid=450885)[0m 'model': {'external_lib': None,
[36m(TaskRunner pid=450885)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=450885)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=450885)[0m 'param_offload': False,
[36m(TaskRunner pid=450885)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=450885)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=450885)[0m 'input_tokenizer': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft',
[36m(TaskRunner pid=450885)[0m 'path': '~/models/FsfairX-LLaMA3-RM-v0.1',
[36m(TaskRunner pid=450885)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=450885)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=450885)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=450885)[0m 'use_shm': False},
[36m(TaskRunner pid=450885)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=450885)[0m 'all_ranks': False,
[36m(TaskRunner pid=450885)[0m 'discrete': False,
[36m(TaskRunner pid=450885)[0m 'ranks': []},
[36m(TaskRunner pid=450885)[0m 'reward_manager': 'naive',
[36m(TaskRunner pid=450885)[0m 'sandbox_fusion': {'max_concurrent': 64,
[36m(TaskRunner pid=450885)[0m 'memory_limit_mb': 1024,
[36m(TaskRunner pid=450885)[0m 'url': None},
[36m(TaskRunner pid=450885)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=450885)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=450885)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=450885)[0m 'trainer': {'balance_batch': True,
[36m(TaskRunner pid=450885)[0m 'controller_nsight_options': {'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=450885)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=450885)[0m 'trace': 'cuda,nvtx,cublas,ucx'},
[36m(TaskRunner pid=450885)[0m 'critic_warmup': 0,
[36m(TaskRunner pid=450885)[0m 'default_hdfs_dir': None,
[36m(TaskRunner pid=450885)[0m 'default_local_dir': '/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints',
[36m(TaskRunner pid=450885)[0m 'del_local_ckpt_after_load': False,
[36m(TaskRunner pid=450885)[0m 'device': 'cuda',
[36m(TaskRunner pid=450885)[0m 'esi_redundant_time': 0,
[36m(TaskRunner pid=450885)[0m 'experiment_name': '0818_rl_alltask_tacc__test_rolloutn16_1gpu_rl',
[36m(TaskRunner pid=450885)[0m 'log_val_generations': 0,
[36m(TaskRunner pid=450885)[0m 'logger': ['console', 'wandb'],
[36m(TaskRunner pid=450885)[0m 'max_actor_ckpt_to_keep': None,
[36m(TaskRunner pid=450885)[0m 'max_critic_ckpt_to_keep': None,
[36m(TaskRunner pid=450885)[0m 'n_gpus_per_node': 1,
[36m(TaskRunner pid=450885)[0m 'nnodes': 1,
[36m(TaskRunner pid=450885)[0m 'profile_steps': None,
[36m(TaskRunner pid=450885)[0m 'project_name': 'jackrl',
[36m(TaskRunner pid=450885)[0m 'ray_wait_register_center_timeout': 300,
[36m(TaskRunner pid=450885)[0m 'resume_from_path': None,
[36m(TaskRunner pid=450885)[0m 'resume_mode': 'disable',
[36m(TaskRunner pid=450885)[0m 'rollout_data_dir': None,
[36m(TaskRunner pid=450885)[0m 'save_freq': 15,
[36m(TaskRunner pid=450885)[0m 'test_freq': 5,
[36m(TaskRunner pid=450885)[0m 'total_epochs': 8,
[36m(TaskRunner pid=450885)[0m 'total_training_steps': None,
[36m(TaskRunner pid=450885)[0m 'val_before_train': False,
[36m(TaskRunner pid=450885)[0m 'val_only': False,
[36m(TaskRunner pid=450885)[0m 'validation_data_dir': None,
[36m(TaskRunner pid=450885)[0m 'worker_nsight_options': {'capture-range': 'cudaProfilerApi',
[36m(TaskRunner pid=450885)[0m 'capture-range-end': None,
[36m(TaskRunner pid=450885)[0m 'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=450885)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=450885)[0m 'kill': 'none',
[36m(TaskRunner pid=450885)[0m 'trace': 'cuda,nvtx,cublas,ucx'}}}
[36m(TaskRunner pid=450885)[0m Registered source: longmult
[36m(TaskRunner pid=450885)[0m Registered source: countdown
[36m(TaskRunner pid=450885)[0m Registered source: gsm8k
[36m(TaskRunner pid=450885)[0m Registered source: arc
[36m(TaskRunner pid=450885)[0m Registered source: arc_challenge
[36m(TaskRunner pid=450885)[0m Registered source: arc_easy
[36m(TaskRunner pid=450885)[0m Registered source: piqa
[36m(TaskRunner pid=450885)[0m Registered source: mmlu
[36m(TaskRunner pid=450885)[0m Registered source: mmlu_pro
[36m(TaskRunner pid=450885)[0m Registered source: csqa
[36m(TaskRunner pid=450885)[0m Registered source: social_iqa
[36m(TaskRunner pid=450885)[0m Registered source: strategy_qa
[36m(TaskRunner pid=450885)[0m Registered source: winogrande
[36m(TaskRunner pid=450885)[0m Registered source: bbh
[36m(TaskRunner pid=450885)[0m using customized reward function 'compute_score' from '/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=450885)[0m using customized reward function 'compute_score' from '/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=450885)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=450885)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=450885)[0m
Generating train split: 1000 examples [00:00, 9751.88 examples/s]
[36m(TaskRunner pid=450885)[0m dataset len: 4000
[36m(TaskRunner pid=450885)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=450885)[0m
Generating train split: 4000 examples [00:00, 10662.90 examples/s]
[36m(TaskRunner pid=450885)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=450885)[0m dataset len: 1000
[36m(TaskRunner pid=450885)[0m Using critic: False
[36m(TaskRunner pid=450885)[0m [validate_config] All configuration checks passed successfully!
[36m(TaskRunner pid=450885)[0m
Generating train split: 1000 examples [00:00, 8649.33 examples/s]
[36m(TaskRunner pid=450885)[0m Size of train dataloader: 15, Size of val dataloader: 1
[36m(TaskRunner pid=450885)[0m Total training steps: 120
[36m(TaskRunner pid=450885)[0m {'b5d105e50b7ed9fe99997185da486466971cd1addbf5b63ff8f9bb04': {'CPU': 71.0,
[36m(TaskRunner pid=450885)[0m 'GPU': 1.0,
[36m(TaskRunner pid=450885)[0m 'accelerator_type:GH200': 1.0,
[36m(TaskRunner pid=450885)[0m 'memory': 159180822937.0,
[36m(TaskRunner pid=450885)[0m 'node:129.114.17.164': 1.0,
[36m(TaskRunner pid=450885)[0m 'node:__internal_head__': 1.0,
[36m(TaskRunner pid=450885)[0m 'object_store_memory': 60055971430.0}}
[36m(TaskRunner pid=450885)[0m ('Resource pool to cls: {<verl.single_controller.ray.base.RayResourcePool '
[36m(TaskRunner pid=450885)[0m "object at 0x400f4900a950>: {'actor_rollout': "
[36m(TaskRunner pid=450885)[0m '<verl.single_controller.ray.base.RayClassWithInitArgs object at '
[36m(TaskRunner pid=450885)[0m '0x400f49065150>}}')
[36m(TaskRunner pid=450885)[0m colocated worker base class <class 'verl.single_controller.base.worker.Worker'>
[36m(TaskRunner pid=450885)[0m DeprecationWarning: `ray.state.available_resources_per_node` is a private attribute and access will be removed in a future Ray version.
[36m(TaskRunner pid=450885)[0m WARNING:2025-08-21 02:31:22,717:Waiting for register center actor bsNqDt_register_center to be ready. Elapsed time: 0 seconds out of 300 seconds.
[36m(pid=451204)[0m /scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/transformers/utils/hub.py:111: FutureWarning: Using `TRANSFORMERS_CACHE` is deprecated and will be removed in v5 of Transformers. Use `HF_HOME` instead.
[36m(pid=451204)[0m warnings.warn(
[36m(WorkerDict pid=451204)[0m Model config after override: Qwen2Config {
[36m(WorkerDict pid=451204)[0m "architectures": [
[36m(WorkerDict pid=451204)[0m "Qwen2ForCausalLM"
[36m(WorkerDict pid=451204)[0m ],
[36m(WorkerDict pid=451204)[0m "attention_dropout": 0.0,
[36m(WorkerDict pid=451204)[0m "eos_token_id": 151645,
[36m(WorkerDict pid=451204)[0m "hidden_act": "silu",
[36m(WorkerDict pid=451204)[0m "hidden_size": 1536,
[36m(WorkerDict pid=451204)[0m "initializer_range": 0.02,
[36m(WorkerDict pid=451204)[0m "intermediate_size": 8960,
[36m(WorkerDict pid=451204)[0m "max_position_embeddings": 32768,
[36m(WorkerDict pid=451204)[0m "max_window_layers": 21,
[36m(WorkerDict pid=451204)[0m "model_type": "qwen2",
[36m(WorkerDict pid=451204)[0m "num_attention_heads": 12,
[36m(WorkerDict pid=451204)[0m "num_hidden_layers": 28,
[36m(WorkerDict pid=451204)[0m "num_key_value_heads": 2,
[36m(WorkerDict pid=451204)[0m "pad_token_id": 151643,
[36m(WorkerDict pid=451204)[0m "rms_norm_eps": 1e-06,
[36m(WorkerDict pid=451204)[0m "rope_scaling": null,
[36m(WorkerDict pid=451204)[0m "rope_theta": 1000000.0,
[36m(WorkerDict pid=451204)[0m "sliding_window": 32768,
[36m(WorkerDict pid=451204)[0m "tie_word_embeddings": true,
[36m(WorkerDict pid=451204)[0m "torch_dtype": "bfloat16",
[36m(WorkerDict pid=451204)[0m "transformers_version": "4.52.3",
[36m(WorkerDict pid=451204)[0m "use_cache": true,
[36m(WorkerDict pid=451204)[0m "use_sliding_window": false,
[36m(WorkerDict pid=451204)[0m "vocab_size": 151936
[36m(WorkerDict pid=451204)[0m }
[36m(WorkerDict pid=451204)[0m
[36m(WorkerDict pid=451204)[0m Flash Attention 2.0 only supports torch.float16 and torch.bfloat16 dtypes, but the current dype in Qwen2ForCausalLM is torch.float32. You should run training or inference using Automatic Mixed-Precision via the `with torch.autocast(device_type='torch_device'):` decorator, or load the model with the `torch_dtype` argument. Example: `model = AutoModel.from_pretrained("openai/whisper-tiny", attn_implementation="flash_attention_2", torch_dtype=torch.float16)`
[36m(WorkerDict pid=451204)[0m You are attempting to use Flash Attention 2.0 with a model not initialized on GPU. Make sure to move the model to GPU after initializing it on CPU with `model.to('cuda')`.
[36m(WorkerDict pid=451204)[0m Monkey patch _flash_attention_forward in transformers.integrations.flash_attention
[36m(WorkerDict pid=451204)[0m Skipping monkey patch for Qwen2ForCausalLM as use_fused_kernels is False or fused_kernels_backend is torch
[36m(WorkerDict pid=451204)[0m Qwen2ForCausalLM contains 1.54B parameters
[36m(WorkerDict pid=451204)[0m wrap_policy: functools.partial(<function _or_policy at 0x400ee4a06e80>, policies=[functools.partial(<function transformer_auto_wrap_policy at 0x400ee4a06d40>, transformer_layer_cls={<class 'transformers.models.qwen2.modeling_qwen2.Qwen2DecoderLayer'>})])
Error executing job with overrides: ['trainer.total_epochs=8', 'actor_rollout_ref.actor.optim.lr=1e-06', 'trainer.save_freq=15', 'trainer.test_freq=5', 'trainer.val_before_train=False', 'algorithm.adv_estimator=grpo', 'actor_rollout_ref.rollout.n=16', 'data.train_batch_size=256', 'actor_rollout_ref.actor.ppo_mini_batch_size=32', 'actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=8', 'actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=16', 'actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=16', 'custom_reward_function.reward_kwargs.format_score_weight=0.0', 'custom_reward_function.reward_kwargs.format_score_v2_weight=0.0', 'custom_reward_function.reward_kwargs.transition_penalty_weight=0.0', 'custom_reward_function.reward_kwargs.reward_min=0.0', 'custom_reward_function.reward_kwargs.reward_max=10.0', 'actor_rollout_ref.model.enable_gradient_checkpointing=True', 'actor_rollout_ref.model.enable_activation_offload=True', 'actor_rollout_ref.rollout.gpu_memory_utilization=0.8', 'actor_rollout_ref.model.use_remove_padding=True', 'actor_rollout_ref.actor.strategy=fsdp2', 'actor_rollout_ref.actor.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.ref.fsdp_config.forward_prefetch=True', 'reward_model.model.fsdp_config.forward_prefetch=True', 'actor_rollout_ref.rollout.tensor_model_parallel_size=1', 'data.max_prompt_length=512', 'data.max_response_length=4096', 'actor_rollout_ref.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft', 'actor_rollout_ref.rollout.dtype=bfloat16', 'critic.optim.lr=1e-05', 'critic.model.path=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/prefetched_models/TAUR_dev__M_skills_in_rl_v2__1e6_all_tasks_sft_sft', 'critic.ppo_micro_batch_size_per_gpu=1', 'algorithm.kl_ctrl.kl_coef=0.001', 'trainer.logger=[console,wandb]', 'trainer.project_name=jackrl', 'trainer.experiment_name=0818_rl_alltask_tacc__test_rolloutn16_1gpu_rl', 'trainer.resume_mode=disable', 'data.train_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/train.parquet', 'data.val_files=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/data/test.parquet', 'custom_reward_function.path=/scratch/10926/yl11330/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py', 'trainer.default_local_dir=/scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints', 'actor_rollout_ref.model.trust_remote_code=True', 'critic.model.trust_remote_code=True', 'trainer.nnodes=1', 'trainer.n_gpus_per_node=1']
[36m(WorkerDict pid=451204)[0m
[36m(WorkerDict pid=451204)[0m c613-021:451204:451204 [0] bootstrap.cc:48 NCCL WARN Bootstrap : no socket interface found
Traceback (most recent call last):
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 39, in main
run_ppo(config)
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 69, in run_ppo
ray.get(runner.run.remote(config))
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/auto_init_hook.py", line 22, in auto_init_wrapper
return fn(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/client_mode_hook.py", line 104, in wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/worker.py", line 2858, in get
values, debugger_breakpoint = worker.get_objects(object_refs, timeout=timeout)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/ray/_private/worker.py", line 958, in get_objects
raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(DistBackendError): [36mray::TaskRunner.run()[39m (pid=450885, ip=129.114.17.164, actor_id=1b985f6a9d7cc0046d5f52de01000000, repr=<main_ppo.TaskRunner object at 0x400f2b1e1dd0>)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 232, in run
trainer.init_workers()
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/trainer/ppo/ray_trainer.py", line 931, in init_workers
self.actor_rollout_wg.init_model()
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 51, in __call__
output = ray.get(output)
^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
ray.exceptions.RayTaskError(DistBackendError): [36mray::WorkerDict.actor_rollout_init_model()[39m (pid=451204, ip=129.114.17.164, actor_id=0d07a09a4af0be8847b07a4501000000, repr=<verl.single_controller.ray.base.WorkerDict object at 0x401128ef7cd0>)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 708, in func
return getattr(self.worker_dict[key], name)(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/single_controller/base/decorator.py", line 549, in inner
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/workers/fsdp_workers.py", line 594, in init_model
) = self._build_model_optimizer(
^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/workers/fsdp_workers.py", line 408, in _build_model_optimizer
fsdp2_load_full_state_dict(actor_module, full_state, fsdp_mesh, cpu_offload)
File "/scratch/10926/yl11330/skill-factory/thirdparty/verl/verl/utils/fsdp_utils.py", line 466, in fsdp2_load_full_state_dict
set_model_state_dict(model, full_state, options=options)
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/checkpoint/state_dict.py", line 1207, in set_model_state_dict
return _load_model_state_dict(model, model_state_dict, info)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/checkpoint/state_dict.py", line 569, in _load_model_state_dict
_broadcast_state_dict(
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/_state_dict_utils.py", line 616, in _broadcast_state_dict
_broadcast_tensors(ret, local_state_dict, keys, device, pg)
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/_state_dict_utils.py", line 535, in _broadcast_tensors
dist.broadcast(tensors[0], src=0, group=pg)
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/c10d_logger.py", line 81, in wrapper
return func(*args, **kwargs)
^^^^^^^^^^^^^^^^^^^^^
File "/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/torch/distributed/distributed_c10d.py", line 2726, in broadcast
work = group.broadcast([tensor], opts)
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
torch.distributed.DistBackendError: NCCL error in: /pytorch/torch/csrc/distributed/c10d/ProcessGroupNCCL.cpp:2588, internal error - please report this issue to the NCCL developers, NCCL version 2.21.5
ncclInternalError: Internal check failed.
Last error:
Bootstrap : no socket interface found
Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.
[INFO] Extracting model from VeRL checkpoint at /scratch/10926/yl11330/skill-factory/workflow_out/0818_rl_alltask_tacc__test_rolloutn16_1gpu/verl/checkpoints
[ERROR] No global_step directories found
EXTRACT OUT: False
[ERROR] Stage error: RuntimeError: Model extraction failed
|
/scratch/10926/yl11330/skill-factory/penv/lib/python3.11/site-packages/huggingface_hub/file_download.py:980: UserWarning: `local_dir_use_symlinks` parameter is deprecated and will be ignored. The process to download files to a local folder has been updated and do not rely on symlinks anymore. You only need to pass a destination folder as`local_dir`.
For more details, check out https://huggingface.co/docs/huggingface_hub/main/en/guides/download#download-files-to-local-folder.
warnings.warn(
Fetching 12 files: 0%| | 0/12 [00:00<?, ?it/s]
Fetching 12 files: 100%|███████████████████████████████████████████████████████████████████████| 12/12 [00:00<00:00, 648.27it/s]
Fetching 12 files: 0%| | 0/12 [00:00<?, ?it/s]
Fetching 12 files: 100%|███████████████████████████████████████████████████████████████████████| 12/12 [00:00<00:00, 701.23it/s]
|
0818_rl_alltask_tacc__test_rolloutn16_1gpu
| 42.748154 | true |
2025-08-23T15:48:09.674537
|
2025-08-23T16:19:04.807833
|
verl_rl
| 1 |
INFO
|
Complete log capture for stage: verl_rl
| "[INFO] Starting stage: VeRL RL training - rl\n[INFO] Data preparation succeeded\n[DEBUG] Running ve(...TRUNCATED) | "\rFetching 12 files: 0%| (...TRUNCATED) |
0818_rl_alltask_tacc__test_rolloutn16_1gpu
| 1,855.133296 | true |
README.md exists but content is empty.
- Downloads last month
- 107