Dataset Viewer
timestamp
string | end_timestamp
string | stage_name
string | stage_number
int64 | level
string | message
string | stdout_content
string | stderr_content
string | experiment_name
string | elapsed_time_seconds
float64 | stage_complete
bool |
|---|---|---|---|---|---|---|---|---|---|---|
2025-08-20T18:08:46.888511
|
2025-08-20T18:20:17.469614
|
verl_rl
| 1 |
INFO
|
Complete log capture for stage: verl_rl
|
[INFO] Starting stage: VeRL RL training - rl
[INFO] Data preparation succeeded
[INFO] Setting up ray cluster
[DEBUG] SLURM cluster info: 4 nodes, 1 GPUs/node
[INFO] Node list: c608-032,c638-[082,091-092]
[DEBUG] Head node: c608-032
[DEBUG] Ray head address: 129.114.17.7:6379
[INFO] Starting Ray head on c608-032...
[INFO] Waiting for head node to initialize...
[DEBUG] Starting 3 worker nodes...
[DEBUG] Starting worker 1: c638-082
[DEBUG] Starting worker 2: c638-091
[DEBUG] Starting worker 3: c638-092
[INFO] Waiting for Ray cluster to stabilize...
[INFO] Connecting to Ray cluster at 129.114.17.7:6379...
[INFO] Ray cluster connected successfully (stats from the connection):
[INFO] Total GPUs: 4.0
[INFO] Available GPUs: 4.0
[INFO] Total CPUs: 288.0
[INFO] SLURM Ray cluster setup completed
[INFO] Starting checkpoint monitoring for intermediate uploads...
[INFO] Intermediate checkpoint upload enabled
[DEBUG] Found 0 global_step directories
[DEBUG] Running verl command:
python -m verl.trainer.main_ppo 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.similarity_penalty_weight=1.0 custom_reward_function.reward_kwargs.sample_correctness_weight=1.0 custom_reward_function.reward_kwargs.sample_count_penalty_weight=1.0 custom_reward_function.reward_kwargs.reward_min=-1.0 custom_reward_function.reward_kwargs.reward_max=10.0 trainer.total_epochs=30 actor_rollout_ref.actor.optim.lr=5e-06 trainer.save_freq=10 trainer.test_freq=10 trainer.val_before_train=False algorithm.adv_estimator=grpo actor_rollout_ref.rollout.n=8 data.train_batch_size=256 actor_rollout_ref.actor.ppo_mini_batch_size=64 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 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/10416/zaynesprague/skill_injection_outputs/sf/grpo_training/exp1/hydra/grpo 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/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft actor_rollout_ref.rollout.dtype=bfloat16 critic.optim.lr=1e-05 critic.model.path=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft critic.ppo_micro_batch_size_per_gpu=1 algorithm.kl_ctrl.kl_coef=0.001 trainer.logger=[console,wandb] trainer.project_name=rl_skills__8_13_25 trainer.experiment_name=VOTING_advanced_grpo_rewards-grpo_adv_rwds_rl trainer.resume_mode=disable data.train_files=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/data/train.parquet data.val_files=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/data/test.parquet custom_reward_function.path=/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py trainer.default_local_dir=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/checkpoints actor_rollout_ref.model.trust_remote_code=True critic.model.trust_remote_code=True trainer.nnodes=4 trainer.n_gpus_per_node=1
2025-08-20 18:10:54,691 INFO worker.py:1554 -- Using address 129.114.17.7:6379 set in the environment variable RAY_ADDRESS
2025-08-20 18:10:54,691 INFO worker.py:1694 -- Connecting to existing Ray cluster at address: 129.114.17.7:6379...
2025-08-20 18:10:54,696 INFO worker.py:1879 -- Connected to Ray cluster. View the dashboard at [1m[32m127.0.0.1:8265 [39m[22m
[36m(TaskRunner pid=2152322)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=2152322)[0m
Generating train split: 1000 examples [00:00, 9422.33 examples/s]
[36m(TaskRunner pid=2152322)[0m
Generating train split: 4000 examples [00:00, 12090.73 examples/s]
[36m(TaskRunner pid=2152322)[0m
Generating train split: 0 examples [00:00, ? examples/s]
[36m(TaskRunner pid=2152322)[0m
Generating train split: 1000 examples [00:00, 11290.19 examples/s]
[36m(TaskRunner pid=2152322)[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=2152322)[0m WARNING:2025-08-20 18:11:09,523:Waiting for register center actor kx9Uxa_register_center to be ready. Elapsed time: 0 seconds out of 300 seconds.
[DEBUG] Found 0 global_step directories
[36m(WorkerDict pid=1619572, ip=129.114.18.209)[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=1619572, ip=129.114.18.209)[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')`.
[DEBUG] Found 0 global_step directories
[36m(TaskRunner pid=2152322)[0m wandb: Currently logged in as: zsprague (ut_nlp_deduce) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin
[36m(WorkerDict pid=3061942, ip=129.114.18.210)[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)`[32m [repeated 3x across cluster][0m
[36m(WorkerDict pid=3061942, ip=129.114.18.210)[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')`.[32m [repeated 3x across cluster][0m
[36m(TaskRunner pid=2152322)[0m wandb: Tracking run with wandb version 0.19.11
[36m(TaskRunner pid=2152322)[0m wandb: Run data is saved locally in /scratch/10416/zaynesprague/skill_factory_dir/skill-factory/wandb/run-20250820_181200-0dge256b
[36m(TaskRunner pid=2152322)[0m wandb: Run `wandb offline` to turn off syncing.
[36m(TaskRunner pid=2152322)[0m wandb: Syncing run VOTING_advanced_grpo_rewards-grpo_adv_rwds_rl
[36m(TaskRunner pid=2152322)[0m wandb: ⭐️ View project at https://wandb.ai/ut_nlp_deduce/rl_skills__8_13_25
[36m(TaskRunner pid=2152322)[0m wandb: 🚀 View run at https://wandb.ai/ut_nlp_deduce/rl_skills__8_13_25/runs/0dge256b
[36m(TaskRunner pid=2152322)[0m
Training Progress: 0%| | 0/450 [00:00<?, ?it/s]
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[DEBUG] Found 0 global_step directories
[36m(TaskRunner pid=2152322)[0m TaskRunner hostname: c608-032.vista.tacc.utexas.edu, PID: 2152322
[36m(TaskRunner pid=2152322)[0m {'actor_rollout_ref': {'actor': {'checkpoint': {'load_contents': ['model',
[36m(TaskRunner pid=2152322)[0m 'optimizer',
[36m(TaskRunner pid=2152322)[0m 'extra'],
[36m(TaskRunner pid=2152322)[0m 'save_contents': ['model',
[36m(TaskRunner pid=2152322)[0m 'optimizer',
[36m(TaskRunner pid=2152322)[0m 'extra']},
[36m(TaskRunner pid=2152322)[0m 'clip_ratio': 0.2,
[36m(TaskRunner pid=2152322)[0m 'clip_ratio_c': 3.0,
[36m(TaskRunner pid=2152322)[0m 'clip_ratio_high': 0.2,
[36m(TaskRunner pid=2152322)[0m 'clip_ratio_low': 0.2,
[36m(TaskRunner pid=2152322)[0m 'entropy_checkpointing': False,
[36m(TaskRunner pid=2152322)[0m 'entropy_coeff': 0,
[36m(TaskRunner pid=2152322)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=2152322)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=2152322)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=2152322)[0m 'offload_policy': False,
[36m(TaskRunner pid=2152322)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=2152322)[0m 'param_offload': False,
[36m(TaskRunner pid=2152322)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=2152322)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=2152322)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=2152322)[0m 'kl_loss_coef': 0.001,
[36m(TaskRunner pid=2152322)[0m 'kl_loss_type': 'low_var_kl',
[36m(TaskRunner pid=2152322)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=2152322)[0m 'optim': {'lr': 5e-06,
[36m(TaskRunner pid=2152322)[0m 'lr_warmup_steps': -1,
[36m(TaskRunner pid=2152322)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=2152322)[0m 'min_lr_ratio': 0.0,
[36m(TaskRunner pid=2152322)[0m 'num_cycles': 0.5,
[36m(TaskRunner pid=2152322)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=2152322)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=2152322)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=2152322)[0m 'policy_loss': {'clip_cov_lb': 1.0,
[36m(TaskRunner pid=2152322)[0m 'clip_cov_ratio': 0.0002,
[36m(TaskRunner pid=2152322)[0m 'clip_cov_ub': 5.0,
[36m(TaskRunner pid=2152322)[0m 'kl_cov_ratio': 0.0002,
[36m(TaskRunner pid=2152322)[0m 'loss_mode': 'vanilla',
[36m(TaskRunner pid=2152322)[0m 'ppo_kl_coef': 0.1},
[36m(TaskRunner pid=2152322)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=2152322)[0m 'ppo_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=2152322)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=2152322)[0m 'ppo_micro_batch_size_per_gpu': 8,
[36m(TaskRunner pid=2152322)[0m 'ppo_mini_batch_size': 64,
[36m(TaskRunner pid=2152322)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=2152322)[0m 'all_ranks': False,
[36m(TaskRunner pid=2152322)[0m 'discrete': False,
[36m(TaskRunner pid=2152322)[0m 'ranks': []},
[36m(TaskRunner pid=2152322)[0m 'shuffle': False,
[36m(TaskRunner pid=2152322)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=2152322)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=2152322)[0m 'use_dynamic_bsz': False,
[36m(TaskRunner pid=2152322)[0m 'use_kl_loss': False,
[36m(TaskRunner pid=2152322)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=2152322)[0m 'hybrid_engine': True,
[36m(TaskRunner pid=2152322)[0m 'model': {'custom_chat_template': None,
[36m(TaskRunner pid=2152322)[0m 'enable_activation_offload': True,
[36m(TaskRunner pid=2152322)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=2152322)[0m 'exclude_modules': None,
[36m(TaskRunner pid=2152322)[0m 'external_lib': None,
[36m(TaskRunner pid=2152322)[0m 'fused_kernel_options': {'impl_backend': 'torch'},
[36m(TaskRunner pid=2152322)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=2152322)[0m 'lora_rank': 0,
[36m(TaskRunner pid=2152322)[0m 'override_config': {},
[36m(TaskRunner pid=2152322)[0m 'path': '/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft',
[36m(TaskRunner pid=2152322)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=2152322)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=2152322)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=2152322)[0m 'use_liger': False,
[36m(TaskRunner pid=2152322)[0m 'use_remove_padding': True,
[36m(TaskRunner pid=2152322)[0m 'use_shm': False},
[36m(TaskRunner pid=2152322)[0m 'ref': {'entropy_checkpointing': False,
[36m(TaskRunner pid=2152322)[0m 'entropy_from_logits_with_chunking': False,
[36m(TaskRunner pid=2152322)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=2152322)[0m 'param_offload': False,
[36m(TaskRunner pid=2152322)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=2152322)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=2152322)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=2152322)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=2152322)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=2152322)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=2152322)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=2152322)[0m 'all_ranks': False,
[36m(TaskRunner pid=2152322)[0m 'discrete': False,
[36m(TaskRunner pid=2152322)[0m 'ranks': []},
[36m(TaskRunner pid=2152322)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=2152322)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=2152322)[0m 'use_torch_compile': True},
[36m(TaskRunner pid=2152322)[0m 'rollout': {'agent': {'custom_async_server': {'name': None,
[36m(TaskRunner pid=2152322)[0m 'path': None},
[36m(TaskRunner pid=2152322)[0m 'num_workers': 8},
[36m(TaskRunner pid=2152322)[0m 'calculate_log_probs': False,
[36m(TaskRunner pid=2152322)[0m 'disable_log_stats': True,
[36m(TaskRunner pid=2152322)[0m 'do_sample': True,
[36m(TaskRunner pid=2152322)[0m 'dtype': 'bfloat16',
[36m(TaskRunner pid=2152322)[0m 'enable_chunked_prefill': True,
[36m(TaskRunner pid=2152322)[0m 'enforce_eager': True,
[36m(TaskRunner pid=2152322)[0m 'engine_kwargs': {'sglang': {'attention_backend': None},
[36m(TaskRunner pid=2152322)[0m 'vllm': {'disable_mm_preprocessor_cache': False,
[36m(TaskRunner pid=2152322)[0m 'swap_space': None}},
[36m(TaskRunner pid=2152322)[0m 'free_cache_engine': True,
[36m(TaskRunner pid=2152322)[0m 'gpu_memory_utilization': 0.8,
[36m(TaskRunner pid=2152322)[0m 'ignore_eos': False,
[36m(TaskRunner pid=2152322)[0m 'layered_summon': False,
[36m(TaskRunner pid=2152322)[0m 'load_format': 'dummy_dtensor',
[36m(TaskRunner pid=2152322)[0m 'log_prob_max_token_len_per_gpu': 16384,
[36m(TaskRunner pid=2152322)[0m 'log_prob_micro_batch_size': None,
[36m(TaskRunner pid=2152322)[0m 'log_prob_micro_batch_size_per_gpu': 16,
[36m(TaskRunner pid=2152322)[0m 'log_prob_use_dynamic_bsz': False,
[36m(TaskRunner pid=2152322)[0m 'max_model_len': None,
[36m(TaskRunner pid=2152322)[0m 'max_num_batched_tokens': 8192,
[36m(TaskRunner pid=2152322)[0m 'max_num_seqs': 1024,
[36m(TaskRunner pid=2152322)[0m 'mode': 'sync',
[36m(TaskRunner pid=2152322)[0m 'multi_stage_wake_up': False,
[36m(TaskRunner pid=2152322)[0m 'multi_turn': {'completion_callback': None,
[36m(TaskRunner pid=2152322)[0m 'enable': False,
[36m(TaskRunner pid=2152322)[0m 'format': 'hermes',
[36m(TaskRunner pid=2152322)[0m 'interaction_config_path': None,
[36m(TaskRunner pid=2152322)[0m 'max_assistant_turns': None,
[36m(TaskRunner pid=2152322)[0m 'max_parallel_calls': 1,
[36m(TaskRunner pid=2152322)[0m 'max_tool_response_length': 256,
[36m(TaskRunner pid=2152322)[0m 'max_user_turns': None,
[36m(TaskRunner pid=2152322)[0m 'tokenization_sanity_check_mode': 'strict',
[36m(TaskRunner pid=2152322)[0m 'tool_config_path': None,
[36m(TaskRunner pid=2152322)[0m 'tool_response_truncate_side': 'middle',
[36m(TaskRunner pid=2152322)[0m 'use_inference_chat_template': False},
[36m(TaskRunner pid=2152322)[0m 'n': 8,
[36m(TaskRunner pid=2152322)[0m 'name': 'vllm',
[36m(TaskRunner pid=2152322)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=2152322)[0m 'all_ranks': False,
[36m(TaskRunner pid=2152322)[0m 'discrete': False,
[36m(TaskRunner pid=2152322)[0m 'ranks': []},
[36m(TaskRunner pid=2152322)[0m 'prompt_length': 512,
[36m(TaskRunner pid=2152322)[0m 'response_length': 4096,
[36m(TaskRunner pid=2152322)[0m 'temperature': 1.0,
[36m(TaskRunner pid=2152322)[0m 'tensor_model_parallel_size': 1,
[36m(TaskRunner pid=2152322)[0m 'top_k': -1,
[36m(TaskRunner pid=2152322)[0m 'top_p': 1,
[36m(TaskRunner pid=2152322)[0m 'val_kwargs': {'do_sample': False,
[36m(TaskRunner pid=2152322)[0m 'n': 1,
[36m(TaskRunner pid=2152322)[0m 'temperature': 0,
[36m(TaskRunner pid=2152322)[0m 'top_k': -1,
[36m(TaskRunner pid=2152322)[0m 'top_p': 1.0}}},
[36m(TaskRunner pid=2152322)[0m 'algorithm': {'_target_': 'verl.trainer.config.AlgoConfig',
[36m(TaskRunner pid=2152322)[0m 'adv_estimator': 'grpo',
[36m(TaskRunner pid=2152322)[0m 'gamma': 1.0,
[36m(TaskRunner pid=2152322)[0m 'kl_ctrl': {'_target_': 'verl.trainer.config.KLControlConfig',
[36m(TaskRunner pid=2152322)[0m 'horizon': 10000,
[36m(TaskRunner pid=2152322)[0m 'kl_coef': 0.001,
[36m(TaskRunner pid=2152322)[0m 'target_kl': 0.1,
[36m(TaskRunner pid=2152322)[0m 'type': 'fixed'},
[36m(TaskRunner pid=2152322)[0m 'kl_penalty': 'kl',
[36m(TaskRunner pid=2152322)[0m 'lam': 1.0,
[36m(TaskRunner pid=2152322)[0m 'norm_adv_by_std_in_grpo': True,
[36m(TaskRunner pid=2152322)[0m 'pf_ppo': {'_target_': 'verl.trainer.config.PFPPOConfig',
[36m(TaskRunner pid=2152322)[0m 'reweight_method': 'pow',
[36m(TaskRunner pid=2152322)[0m 'weight_pow': 2.0},
[36m(TaskRunner pid=2152322)[0m 'use_kl_in_reward': False,
[36m(TaskRunner pid=2152322)[0m 'use_pf_ppo': False},
[36m(TaskRunner pid=2152322)[0m 'critic': {'checkpoint': {'load_contents': ['model', 'optimizer', 'extra'],
[36m(TaskRunner pid=2152322)[0m 'save_contents': ['model', 'optimizer', 'extra']},
[36m(TaskRunner pid=2152322)[0m 'cliprange_value': 0.5,
[36m(TaskRunner pid=2152322)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=2152322)[0m 'forward_micro_batch_size': None,
[36m(TaskRunner pid=2152322)[0m 'forward_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=2152322)[0m 'grad_clip': 1.0,
[36m(TaskRunner pid=2152322)[0m 'loss_agg_mode': 'token-mean',
[36m(TaskRunner pid=2152322)[0m 'model': {'enable_activation_offload': False,
[36m(TaskRunner pid=2152322)[0m 'enable_gradient_checkpointing': True,
[36m(TaskRunner pid=2152322)[0m 'external_lib': None,
[36m(TaskRunner pid=2152322)[0m 'fsdp_config': {'forward_prefetch': False,
[36m(TaskRunner pid=2152322)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=2152322)[0m 'offload_policy': False,
[36m(TaskRunner pid=2152322)[0m 'optimizer_offload': False,
[36m(TaskRunner pid=2152322)[0m 'param_offload': False,
[36m(TaskRunner pid=2152322)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=2152322)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=2152322)[0m 'lora_alpha': 16,
[36m(TaskRunner pid=2152322)[0m 'lora_rank': 0,
[36m(TaskRunner pid=2152322)[0m 'override_config': {},
[36m(TaskRunner pid=2152322)[0m 'path': '/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft',
[36m(TaskRunner pid=2152322)[0m 'target_modules': 'all-linear',
[36m(TaskRunner pid=2152322)[0m 'tokenizer_path': '/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft',
[36m(TaskRunner pid=2152322)[0m 'trust_remote_code': True,
[36m(TaskRunner pid=2152322)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=2152322)[0m 'use_shm': False},
[36m(TaskRunner pid=2152322)[0m 'optim': {'lr': 1e-05,
[36m(TaskRunner pid=2152322)[0m 'lr_warmup_steps_ratio': 0.0,
[36m(TaskRunner pid=2152322)[0m 'min_lr_ratio': None,
[36m(TaskRunner pid=2152322)[0m 'total_training_steps': -1,
[36m(TaskRunner pid=2152322)[0m 'warmup_style': 'constant',
[36m(TaskRunner pid=2152322)[0m 'weight_decay': 0.01},
[36m(TaskRunner pid=2152322)[0m 'ppo_epochs': 1,
[36m(TaskRunner pid=2152322)[0m 'ppo_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=2152322)[0m 'ppo_micro_batch_size': None,
[36m(TaskRunner pid=2152322)[0m 'ppo_micro_batch_size_per_gpu': 1,
[36m(TaskRunner pid=2152322)[0m 'ppo_mini_batch_size': 64,
[36m(TaskRunner pid=2152322)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=2152322)[0m 'all_ranks': False,
[36m(TaskRunner pid=2152322)[0m 'discrete': False,
[36m(TaskRunner pid=2152322)[0m 'ranks': []},
[36m(TaskRunner pid=2152322)[0m 'rollout_n': 8,
[36m(TaskRunner pid=2152322)[0m 'shuffle': False,
[36m(TaskRunner pid=2152322)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=2152322)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=2152322)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=2152322)[0m 'custom_reward_function': {'name': 'compute_score',
[36m(TaskRunner pid=2152322)[0m 'path': '/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py',
[36m(TaskRunner pid=2152322)[0m 'reward_kwargs': {'format_score_v2_weight': 0.0,
[36m(TaskRunner pid=2152322)[0m 'format_score_weight': 0.0,
[36m(TaskRunner pid=2152322)[0m 'reward_max': 10.0,
[36m(TaskRunner pid=2152322)[0m 'reward_min': -1.0,
[36m(TaskRunner pid=2152322)[0m 'sample_correctness_weight': 1.0,
[36m(TaskRunner pid=2152322)[0m 'sample_count_penalty_weight': 1.0,
[36m(TaskRunner pid=2152322)[0m 'similarity_penalty_weight': 1.0,
[36m(TaskRunner pid=2152322)[0m 'transition_penalty_weight': 0.0}},
[36m(TaskRunner pid=2152322)[0m 'data': {'custom_cls': {'name': None, 'path': None},
[36m(TaskRunner pid=2152322)[0m 'dataloader_num_workers': 8,
[36m(TaskRunner pid=2152322)[0m 'filter_overlong_prompts': False,
[36m(TaskRunner pid=2152322)[0m 'filter_overlong_prompts_workers': 1,
[36m(TaskRunner pid=2152322)[0m 'image_key': 'images',
[36m(TaskRunner pid=2152322)[0m 'max_prompt_length': 512,
[36m(TaskRunner pid=2152322)[0m 'max_response_length': 4096,
[36m(TaskRunner pid=2152322)[0m 'prompt_key': 'prompt',
[36m(TaskRunner pid=2152322)[0m 'return_full_prompt': False,
[36m(TaskRunner pid=2152322)[0m 'return_multi_modal_inputs': True,
[36m(TaskRunner pid=2152322)[0m 'return_raw_chat': False,
[36m(TaskRunner pid=2152322)[0m 'return_raw_input_ids': False,
[36m(TaskRunner pid=2152322)[0m 'reward_fn_key': 'data_source',
[36m(TaskRunner pid=2152322)[0m 'sampler': {'class_name': None, 'class_path': None},
[36m(TaskRunner pid=2152322)[0m 'shuffle': True,
[36m(TaskRunner pid=2152322)[0m 'tokenizer': None,
[36m(TaskRunner pid=2152322)[0m 'train_batch_size': 256,
[36m(TaskRunner pid=2152322)[0m 'train_files': '/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/data/train.parquet',
[36m(TaskRunner pid=2152322)[0m 'truncation': 'error',
[36m(TaskRunner pid=2152322)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=2152322)[0m 'use_shm': False,
[36m(TaskRunner pid=2152322)[0m 'val_batch_size': None,
[36m(TaskRunner pid=2152322)[0m 'val_files': '/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/data/test.parquet',
[36m(TaskRunner pid=2152322)[0m 'validation_shuffle': False,
[36m(TaskRunner pid=2152322)[0m 'video_key': 'videos'},
[36m(TaskRunner pid=2152322)[0m 'ray_init': {'num_cpus': None, 'timeline_json_file': None},
[36m(TaskRunner pid=2152322)[0m 'reward_model': {'enable': False,
[36m(TaskRunner pid=2152322)[0m 'forward_max_token_len_per_gpu': 32768,
[36m(TaskRunner pid=2152322)[0m 'launch_reward_fn_async': False,
[36m(TaskRunner pid=2152322)[0m 'max_length': None,
[36m(TaskRunner pid=2152322)[0m 'micro_batch_size': None,
[36m(TaskRunner pid=2152322)[0m 'micro_batch_size_per_gpu': None,
[36m(TaskRunner pid=2152322)[0m 'model': {'external_lib': None,
[36m(TaskRunner pid=2152322)[0m 'fsdp_config': {'forward_prefetch': True,
[36m(TaskRunner pid=2152322)[0m 'fsdp_size': -1,
[36m(TaskRunner pid=2152322)[0m 'param_offload': False,
[36m(TaskRunner pid=2152322)[0m 'reshard_after_forward': True,
[36m(TaskRunner pid=2152322)[0m 'wrap_policy': {'min_num_params': 0}},
[36m(TaskRunner pid=2152322)[0m 'input_tokenizer': '/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft',
[36m(TaskRunner pid=2152322)[0m 'path': '~/models/FsfairX-LLaMA3-RM-v0.1',
[36m(TaskRunner pid=2152322)[0m 'trust_remote_code': False,
[36m(TaskRunner pid=2152322)[0m 'use_fused_kernels': False,
[36m(TaskRunner pid=2152322)[0m 'use_remove_padding': False,
[36m(TaskRunner pid=2152322)[0m 'use_shm': False},
[36m(TaskRunner pid=2152322)[0m 'profiler': {'_target_': 'verl.utils.profiler.ProfilerConfig',
[36m(TaskRunner pid=2152322)[0m 'all_ranks': False,
[36m(TaskRunner pid=2152322)[0m 'discrete': False,
[36m(TaskRunner pid=2152322)[0m 'ranks': []},
[36m(TaskRunner pid=2152322)[0m 'reward_manager': 'naive',
[36m(TaskRunner pid=2152322)[0m 'sandbox_fusion': {'max_concurrent': 64,
[36m(TaskRunner pid=2152322)[0m 'memory_limit_mb': 1024,
[36m(TaskRunner pid=2152322)[0m 'url': None},
[36m(TaskRunner pid=2152322)[0m 'strategy': 'fsdp2',
[36m(TaskRunner pid=2152322)[0m 'ulysses_sequence_parallel_size': 1,
[36m(TaskRunner pid=2152322)[0m 'use_dynamic_bsz': False},
[36m(TaskRunner pid=2152322)[0m 'trainer': {'balance_batch': True,
[36m(TaskRunner pid=2152322)[0m 'controller_nsight_options': {'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=2152322)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=2152322)[0m 'trace': 'cuda,nvtx,cublas,ucx'},
[36m(TaskRunner pid=2152322)[0m 'critic_warmup': 0,
[36m(TaskRunner pid=2152322)[0m 'default_hdfs_dir': None,
[36m(TaskRunner pid=2152322)[0m 'default_local_dir': '/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/checkpoints',
[36m(TaskRunner pid=2152322)[0m 'del_local_ckpt_after_load': False,
[36m(TaskRunner pid=2152322)[0m 'device': 'cuda',
[36m(TaskRunner pid=2152322)[0m 'esi_redundant_time': 0,
[36m(TaskRunner pid=2152322)[0m 'experiment_name': 'VOTING_advanced_grpo_rewards-grpo_adv_rwds_rl',
[36m(TaskRunner pid=2152322)[0m 'log_val_generations': 0,
[36m(TaskRunner pid=2152322)[0m 'logger': ['console', 'wandb'],
[36m(TaskRunner pid=2152322)[0m 'max_actor_ckpt_to_keep': None,
[36m(TaskRunner pid=2152322)[0m 'max_critic_ckpt_to_keep': None,
[36m(TaskRunner pid=2152322)[0m 'n_gpus_per_node': 1,
[36m(TaskRunner pid=2152322)[0m 'nnodes': 4,
[36m(TaskRunner pid=2152322)[0m 'profile_steps': None,
[36m(TaskRunner pid=2152322)[0m 'project_name': 'rl_skills__8_13_25',
[36m(TaskRunner pid=2152322)[0m 'ray_wait_register_center_timeout': 300,
[36m(TaskRunner pid=2152322)[0m 'resume_from_path': None,
[36m(TaskRunner pid=2152322)[0m 'resume_mode': 'disable',
[36m(TaskRunner pid=2152322)[0m 'rollout_data_dir': None,
[36m(TaskRunner pid=2152322)[0m 'save_freq': 10,
[36m(TaskRunner pid=2152322)[0m 'test_freq': 10,
[36m(TaskRunner pid=2152322)[0m 'total_epochs': 30,
[36m(TaskRunner pid=2152322)[0m 'total_training_steps': None,
[36m(TaskRunner pid=2152322)[0m 'val_before_train': False,
[36m(TaskRunner pid=2152322)[0m 'val_only': False,
[36m(TaskRunner pid=2152322)[0m 'validation_data_dir': None,
[36m(TaskRunner pid=2152322)[0m 'worker_nsight_options': {'capture-range': 'cudaProfilerApi',
[36m(TaskRunner pid=2152322)[0m 'capture-range-end': None,
[36m(TaskRunner pid=2152322)[0m 'cuda-graph-trace': 'graph',
[36m(TaskRunner pid=2152322)[0m 'cuda-memory-usage': 'true',
[36m(TaskRunner pid=2152322)[0m 'kill': 'none',
[36m(TaskRunner pid=2152322)[0m 'trace': 'cuda,nvtx,cublas,ucx'}}}
[36m(TaskRunner pid=2152322)[0m Registered source: longmult
[36m(TaskRunner pid=2152322)[0m Registered source: countdown
[36m(TaskRunner pid=2152322)[0m Registered source: gsm8k
[36m(TaskRunner pid=2152322)[0m Registered source: arc
[36m(TaskRunner pid=2152322)[0m Registered source: arc_challenge
[36m(TaskRunner pid=2152322)[0m Registered source: arc_easy
[36m(TaskRunner pid=2152322)[0m Registered source: piqa
[36m(TaskRunner pid=2152322)[0m Registered source: mmlu
[36m(TaskRunner pid=2152322)[0m Registered source: mmlu_pro
[36m(TaskRunner pid=2152322)[0m Registered source: csqa
[36m(TaskRunner pid=2152322)[0m Registered source: social_iqa
[36m(TaskRunner pid=2152322)[0m Registered source: strategy_qa
[36m(TaskRunner pid=2152322)[0m Registered source: winogrande
[36m(TaskRunner pid=2152322)[0m Registered source: bbh
[36m(TaskRunner pid=2152322)[0m using customized reward function 'compute_score' from '/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=2152322)[0m using customized reward function 'compute_score' from '/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py'
[36m(TaskRunner pid=2152322)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=2152322)[0m dataset len: 4000
[36m(TaskRunner pid=2152322)[0m Using dataset class: RLHFDataset
[36m(TaskRunner pid=2152322)[0m dataset len: 1000
[36m(TaskRunner pid=2152322)[0m Using critic: False
[36m(TaskRunner pid=2152322)[0m [validate_config] All configuration checks passed successfully!
[36m(TaskRunner pid=2152322)[0m Size of train dataloader: 15, Size of val dataloader: 1
[36m(TaskRunner pid=2152322)[0m Total training steps: 450
[36m(TaskRunner pid=2152322)[0m {'4499579ce8b9c8ea0a34fbdb005c541afa0e0a8211c9de26cc5218e8': {'CPU': 72.0,
[36m(TaskRunner pid=2152322)[0m 'GPU': 1.0,
[36m(TaskRunner pid=2152322)[0m 'accelerator_type:GH200': 1.0,
[36m(TaskRunner pid=2152322)[0m 'memory': 160702703206.0,
[36m(TaskRunner pid=2152322)[0m 'node:129.114.18.209': 1.0,
[36m(TaskRunner pid=2152322)[0m 'object_store_memory': 60056033689.0},
[36m(TaskRunner pid=2152322)[0m '6f003a8d18b4f1740c4e8ae45740efe6f39f67a5fc884ba5173230ea': {'CPU': 71.0,
[36m(TaskRunner pid=2152322)[0m 'GPU': 1.0,
[36m(TaskRunner pid=2152322)[0m 'accelerator_type:GH200': 1.0,
[36m(TaskRunner pid=2152322)[0m 'memory': 159528425881.0,
[36m(TaskRunner pid=2152322)[0m 'node:129.114.17.7': 1.0,
[36m(TaskRunner pid=2152322)[0m 'node:__internal_head__': 1.0,
[36m(TaskRunner pid=2152322)[0m 'object_store_memory': 60055971430.0},
[36m(TaskRunner pid=2152322)[0m '7106ae01328147693283e38c0fe8f364c51fc81151b7a5d64fbe79d6': {'CPU': 72.0,
[36m(TaskRunner pid=2152322)[0m 'GPU': 1.0,
[36m(TaskRunner pid=2152322)[0m 'accelerator_type:GH200': 1.0,
[36m(TaskRunner pid=2152322)[0m 'memory': 160679306854.0,
[36m(TaskRunner pid=2152322)[0m 'node:129.114.18.210': 1.0,
[36m(TaskRunner pid=2152322)[0m 'object_store_memory': 60056033689.0},
[36m(TaskRunner pid=2152322)[0m '78f9463a3cc562f235a681ec7ad5e25f94f61b2f4b92af9d6c6dde5e': {'CPU': 72.0,
[36m(TaskRunner pid=2152322)[0m 'GPU': 1.0,
[36m(TaskRunner pid=2152322)[0m 'accelerator_type:GH200': 1.0,
[36m(TaskRunner pid=2152322)[0m 'memory': 160690644582.0,
[36m(TaskRunner pid=2152322)[0m 'node:129.114.18.211': 1.0,
[36m(TaskRunner pid=2152322)[0m 'object_store_memory': 60056033689.0}}
[36m(TaskRunner pid=2152322)[0m ('Resource pool to cls: {<verl.single_controller.ray.base.RayResourcePool '
[36m(TaskRunner pid=2152322)[0m "object at 0x400f5cd964d0>: {'actor_rollout': "
[36m(TaskRunner pid=2152322)[0m '<verl.single_controller.ray.base.RayClassWithInitArgs object at '
[36m(TaskRunner pid=2152322)[0m '0x400f5cd96500>}}')
[36m(TaskRunner pid=2152322)[0m colocated worker base class <class 'verl.single_controller.base.worker.Worker'>
[36m(WorkerDict pid=2153809)[0m Model config after override: Qwen2Config {
[36m(WorkerDict pid=2153809)[0m "_name_or_path": "/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft",
[36m(WorkerDict pid=2153809)[0m "architectures": [
[36m(WorkerDict pid=2153809)[0m "Qwen2ForCausalLM"
[36m(WorkerDict pid=2153809)[0m ],
[36m(WorkerDict pid=2153809)[0m "attention_dropout": 0.0,
[36m(WorkerDict pid=2153809)[0m "eos_token_id": 151645,
[36m(WorkerDict pid=2153809)[0m "hidden_act": "silu",
[36m(WorkerDict pid=2153809)[0m "hidden_size": 1536,
[36m(WorkerDict pid=2153809)[0m "initializer_range": 0.02,
[36m(WorkerDict pid=2153809)[0m "intermediate_size": 8960,
[36m(WorkerDict pid=2153809)[0m "max_position_embeddings": 32768,
[36m(WorkerDict pid=2153809)[0m "max_window_layers": 21,
[36m(WorkerDict pid=2153809)[0m "model_type": "qwen2",
[36m(WorkerDict pid=2153809)[0m "num_attention_heads": 12,
[36m(WorkerDict pid=2153809)[0m "num_hidden_layers": 28,
[36m(WorkerDict pid=2153809)[0m "num_key_value_heads": 2,
[36m(WorkerDict pid=2153809)[0m "pad_token_id": 151643,
[36m(WorkerDict pid=2153809)[0m "rms_norm_eps": 1e-06,
[36m(WorkerDict pid=2153809)[0m "rope_scaling": null,
[36m(WorkerDict pid=2153809)[0m "rope_theta": 1000000.0,
[36m(WorkerDict pid=2153809)[0m "sliding_window": 32768,
[36m(WorkerDict pid=2153809)[0m "tie_word_embeddings": true,
[36m(WorkerDict pid=2153809)[0m "torch_dtype": "bfloat16",
[36m(WorkerDict pid=2153809)[0m "transformers_version": "4.49.0",
[36m(WorkerDict pid=2153809)[0m "use_cache": true,
[36m(WorkerDict pid=2153809)[0m "use_sliding_window": false,
[36m(WorkerDict pid=2153809)[0m "vocab_size": 151936
[36m(WorkerDict pid=2153809)[0m }
[36m(WorkerDict pid=2153809)[0m
[36m(WorkerDict pid=269696, ip=129.114.18.211)[0m Monkey patch state_dict in AutoModelForCausalLMWithValueHead.
[36m(WorkerDict pid=269696, ip=129.114.18.211)[0m Monkey patch _flash_attention_forward in transformers.integrations.flash_attention
[36m(WorkerDict pid=269696, ip=129.114.18.211)[0m Skipping monkey patch for Qwen2ForCausalLM as use_fused_kernels is False or fused_kernels_backend is torch
[36m(WorkerDict pid=2153809)[0m Qwen2ForCausalLM contains 1.54B parameters
[36m(WorkerDict pid=2153809)[0m wrap_policy: functools.partial(<function _or_policy at 0x400ebe83d990>, policies=[functools.partial(<function transformer_auto_wrap_policy at 0x400ebe83d870>, transformer_layer_cls={<class 'transformers.models.qwen2.modeling_qwen2.Qwen2DecoderLayer'>})])
[36m(WorkerDict pid=2153809)[0m NCCL version 2.21.5+cuda12.6
[36m(WorkerDict pid=2153809)[0m Total steps: 450, num_warmup_steps: 0
[36m(WorkerDict pid=2153809)[0m Actor use_remove_padding=True
[36m(WorkerDict pid=2153809)[0m Actor use_fused_kernels=False
[36m(WorkerDict pid=1619572, ip=129.114.18.209)[0m Monkey patch state_dict in AutoModelForCausalLMWithValueHead. [32m [repeated 3x 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=1619572, ip=129.114.18.209)[0m Monkey patch _flash_attention_forward in transformers.integrations.flash_attention[32m [repeated 3x across cluster][0m
[36m(WorkerDict pid=1619572, ip=129.114.18.209)[0m Skipping monkey patch for Qwen2ForCausalLM as use_fused_kernels is False or fused_kernels_backend is torch[32m [repeated 3x across cluster][0m
[36m(WorkerDict pid=2153809)[0m WARNING 08-20 18:11:53 [cuda.py:93] To see benefits of async output processing, enable CUDA graph. Since, enforce-eager is enabled, async output processor cannot be used
[36m(WorkerDict pid=2153809)[0m WARNING 08-20 18:11:55 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in <vllm.v1.worker.gpu_worker.Worker object at 0x40118e1a4dc0>
[36m(WorkerDict pid=269696, ip=129.114.18.211)[0m WARNING 08-20 18:11:57 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer.
[36m(TaskRunner pid=2152322)[0m
[36m(TaskRunner pid=2152322)[0m ======== RayPPOTrainer Fields START ========
[36m(TaskRunner pid=2152322)[0m processor: None
[36m(TaskRunner pid=2152322)[0m reward_fn: <verl.workers.reward_manager.naive.NaiveRewardManager object at 0x400f5a61f970>
[36m(TaskRunner pid=2152322)[0m val_reward_fn: <verl.workers.reward_manager.naive.NaiveRewardManager object at 0x400f5a61fd00>
[36m(TaskRunner pid=2152322)[0m use_reference_policy: False
[36m(TaskRunner pid=2152322)[0m use_rm: False
[36m(TaskRunner pid=2152322)[0m device_name: cuda
[36m(TaskRunner pid=2152322)[0m validation_generations_logger: ValidationGenerationsLogger()
[36m(TaskRunner pid=2152322)[0m ref_in_actor: False
[36m(TaskRunner pid=2152322)[0m config.algorithm.adv_estimator: grpo
[36m(TaskRunner pid=2152322)[0m use_critic False
[36m(TaskRunner pid=2152322)[0m ======== RayPPOTrainer Fields END ========
[36m(TaskRunner pid=2152322)[0m
[36m(WorkerDict pid=2153809)[0m vLLM Rollout kwargs: {'n': 1, 'logprobs': 0, 'max_tokens': 4096, 'detokenize': False, 'temperature': 1.0, 'top_k': -1, 'top_p': 1, 'ignore_eos': False}
Error executing job with overrides: ['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.similarity_penalty_weight=1.0', 'custom_reward_function.reward_kwargs.sample_correctness_weight=1.0', 'custom_reward_function.reward_kwargs.sample_count_penalty_weight=1.0', 'custom_reward_function.reward_kwargs.reward_min=-1.0', 'custom_reward_function.reward_kwargs.reward_max=10.0', 'trainer.total_epochs=30', 'actor_rollout_ref.actor.optim.lr=5e-06', 'trainer.save_freq=10', 'trainer.test_freq=10', 'trainer.val_before_train=False', 'algorithm.adv_estimator=grpo', 'actor_rollout_ref.rollout.n=8', 'data.train_batch_size=256', 'actor_rollout_ref.actor.ppo_mini_batch_size=64', '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', '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/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft', 'actor_rollout_ref.rollout.dtype=bfloat16', 'critic.optim.lr=1e-05', 'critic.model.path=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/prefetched_models/TAUR_dev__M_voting_setup1_1epch_1e6_all_tasks_only_sft_sft', 'critic.ppo_micro_batch_size_per_gpu=1', 'algorithm.kl_ctrl.kl_coef=0.001', 'trainer.logger=[console,wandb]', 'trainer.project_name=rl_skills__8_13_25', 'trainer.experiment_name=VOTING_advanced_grpo_rewards-grpo_adv_rwds_rl', 'trainer.resume_mode=disable', 'data.train_files=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/data/train.parquet', 'data.val_files=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/data/test.parquet', 'custom_reward_function.path=/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/sf_scripts/skill_factory_rewards.py', 'trainer.default_local_dir=/scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/checkpoints', 'actor_rollout_ref.model.trust_remote_code=True', 'critic.model.trust_remote_code=True', 'trainer.nnodes=4', 'trainer.n_gpus_per_node=1']
Traceback (most recent call last):
File "/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 39, in main
run_ppo(config)
File "/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 69, in run_ppo
ray.get(runner.run.remote(config))
File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/_private/auto_init_hook.py", line 21, in auto_init_wrapper
return fn(*args, **kwargs)
File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/_private/client_mode_hook.py", line 103, in wrapper
return func(*args, **kwargs)
File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/_private/worker.py", line 2822, in get
values, debugger_breakpoint = worker.get_objects(object_refs, timeout=timeout)
File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/_private/worker.py", line 930, in get_objects
raise value.as_instanceof_cause()
ray.exceptions.RayTaskError(RaySystemError): [36mray::TaskRunner.run()[39m (pid=2152322, ip=129.114.17.7, actor_id=61742b1e6d6c5e6e5576943e02000000, repr=<main_ppo.TaskRunner object at 0x400f42b14f70>)
File "/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/verl/trainer/main_ppo.py", line 234, in run
trainer.fit()
File "/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/verl/trainer/ppo/ray_trainer.py", line 1244, in fit
old_log_prob = self.actor_rollout_wg.compute_log_prob(batch)
File "/scratch/10416/zaynesprague/skill_factory_dir/skill-factory/thirdparty/verl/verl/single_controller/ray/base.py", line 51, in __call__
output = ray.get(output)
ray.exceptions.RaySystemError: System error: Failed to unpickle serialized exception
traceback: Traceback (most recent call last):
File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/exceptions.py", line 51, in from_ray_exception
return pickle.loads(ray_exception.serialized_exception)
TypeError: BackendCompilerFailed.__init__() missing 1 required positional argument: 'inner_exception'
The above exception was the direct cause of the following exception:
[36mray::TaskRunner.run()[39m (pid=2152322, ip=129.114.17.7, actor_id=61742b1e6d6c5e6e5576943e02000000, repr=<main_ppo.TaskRunner object at 0x400f42b14f70>)
File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/exceptions.py", line 45, in from_bytes
return RayError.from_ray_exception(ray_exception)
File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/exceptions.py", line 54, in from_ray_exception
raise RuntimeError(msg) from e
RuntimeError: Failed to unpickle serialized exception
Set the environment variable HYDRA_FULL_ERROR=1 for a complete stack trace.
[36m(WorkerDict pid=269696, ip=129.114.18.211)[0m nvc-Error-Unknown switch: -Wno-psabi
[36m(TaskRunner pid=2152322)[0m Failed to unpickle serialized exception
[36m(TaskRunner pid=2152322)[0m Traceback (most recent call last):
[36m(TaskRunner pid=2152322)[0m File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/exceptions.py", line 51, in from_ray_exception
[36m(TaskRunner pid=2152322)[0m return pickle.loads(ray_exception.serialized_exception)
[36m(TaskRunner pid=2152322)[0m TypeError: BackendCompilerFailed.__init__() missing 1 required positional argument: 'inner_exception'
[36m(TaskRunner pid=2152322)[0m
[36m(TaskRunner pid=2152322)[0m The above exception was the direct cause of the following exception:
[36m(TaskRunner pid=2152322)[0m
[36m(TaskRunner pid=2152322)[0m Traceback (most recent call last):
[36m(TaskRunner pid=2152322)[0m File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/_private/serialization.py", line 460, in deserialize_objects
[36m(TaskRunner pid=2152322)[0m obj = self._deserialize_object(data, metadata, object_ref)
[36m(TaskRunner pid=2152322)[0m File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/_private/serialization.py", line 342, in _deserialize_object
[36m(TaskRunner pid=2152322)[0m return RayError.from_bytes(obj)
[36m(TaskRunner pid=2152322)[0m File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/exceptions.py", line 45, in from_bytes
[36m(TaskRunner pid=2152322)[0m return RayError.from_ray_exception(ray_exception)
[36m(TaskRunner pid=2152322)[0m File "/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/site-packages/ray/exceptions.py", line 54, in from_ray_exception
[36m(TaskRunner pid=2152322)[0m raise RuntimeError(msg) from e
[36m(TaskRunner pid=2152322)[0m RuntimeError: Failed to unpickle serialized exception
[36m(WorkerDict pid=3061942, ip=129.114.18.210)[0m nvc-Error-Unknown switch: -Wno-psabi[32m [repeated 3x across cluster][0m
[36m(WorkerDict pid=3061942, ip=129.114.18.210)[0m WARNING 08-20 18:11:55 [cuda.py:93] To see benefits of async output processing, enable CUDA graph. Since, enforce-eager is enabled, async output processor cannot be used[32m [repeated 3x across cluster][0m
[36m(WorkerDict pid=3061942, ip=129.114.18.210)[0m WARNING 08-20 18:11:56 [utils.py:2522] Methods determine_num_available_blocks,device_config,get_cache_block_size_bytes,initialize_cache not implemented in <vllm.v1.worker.gpu_worker.Worker object at 0x40117bdea230>[32m [repeated 3x across cluster][0m
[36m(WorkerDict pid=1619572, ip=129.114.18.209)[0m WARNING 08-20 18:11:57 [topk_topp_sampler.py:69] FlashInfer is not available. Falling back to the PyTorch-native implementation of top-p & top-k sampling. For the best performance, please install FlashInfer.[32m [repeated 3x across cluster][0m
[36m(WorkerDict pid=269696, ip=129.114.18.211)[0m vLLM Rollout kwargs: {'n': 1, 'logprobs': 0, 'max_tokens': 4096, 'detokenize': False, 'temperature': 1.0, 'top_k': -1, 'top_p': 1, 'ignore_eos': False}[32m [repeated 3x across cluster][0m
[INFO] Extracting model from VeRL checkpoint at /scratch/10416/zaynesprague/skill_inject_outputs/sf_experiments/skills_in_rl/grpo_lexical_sim/all_tasks/VOTING_advanced_grpo_rewards/verl/checkpoints
[ERROR] No global_step directories found
EXTRACT OUT: False
[ERROR] Stage error: RuntimeError: Model extraction failed
|
Fetching 13 files: 0%| | 0/13 [00:00<?, ?it/s]/work/10416/zaynesprague/anaconda3/envs/verl2/lib/python3.10/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 13 files: 8%|▊ | 1/13 [00:00<00:03, 3.14it/s]
Fetching 13 files: 69%|██████▉ | 9/13 [00:53<00:24, 6.12s/it]
Fetching 13 files: 100%|██████████| 13/13 [00:53<00:00, 4.11s/it]
Fetching 13 files: 0%| | 0/13 [00:00<?, ?it/s]
Fetching 13 files: 100%|██████████| 13/13 [00:00<00:00, 175.40it/s]
2025-08-20 18:10:45,469 INFO worker.py:1694 -- Connecting to existing Ray cluster at address: 129.114.17.7:6379...
2025-08-20 18:10:45,474 INFO worker.py:1879 -- Connected to Ray cluster. View the dashboard at [1m[32m127.0.0.1:8265 [39m[22m
|
VOTING_advanced_grpo_rewards-grpo_adv_rwds
| 690.581103 | true |
README.md exists but content is empty.
- Downloads last month
- 2