diff --git a/outputs/2026-06-01/23-08-51/.hydra/config.yaml b/outputs/2026-06-01/23-08-51/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..bf5373dad6ad7b8941dab9abd4113694a84b8766 --- /dev/null +++ b/outputs/2026-06-01/23-08-51/.hydra/config.yaml @@ -0,0 +1,1005 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 400 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 1000 + project_name: ragen + experiment_name: sudoku_3b_baseline + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 20 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 20 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/share/model/Qwen/Qwen2.5-3B-Instruct +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: -1 + batch_adjust_mode: copy + max_turn: 20 + action_sep: '||' + max_actions_per_turn: 3 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - SimpleSudoku + n_groups: + - 8 + val: + env_groups: 32 + group_size: 16 + env_configs: + tags: + - SimpleSudoku + n_groups: + - 32 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-07/23-18-01/.hydra/hydra.yaml b/outputs/2026-06-07/23-18-01/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..127fd55999084586b6a999be68a1cd1e9daba4f4 --- /dev/null +++ b/outputs/2026-06-07/23-18-01/.hydra/hydra.yaml @@ -0,0 +1,178 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: + - pkg://verl.trainer.config + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_sas + - trainer.experiment_name=sokoban_3b_box2_withthink_sas + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_sas,trainer.experiment_name=sokoban_3b_box2_withthink_sas,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _2_sokoban + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: verl.trainer.config + schema: pkg + provider: hydra.searchpath in main + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-07/23-18-01 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-09/00-36-11/.hydra/hydra.yaml b/outputs/2026-06-09/00-36-11/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f564182f6e4c5275617856fc08cf527e045e11d1 --- /dev/null +++ b/outputs/2026-06-09/00-36-11/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sas + - custom_envs.rubikscube.env_config.scramble_depth=1 + - trainer.experiment_name=rubikscube1_withthink_sas + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=1,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube1_withthink_sas,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-09/00-36-11 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-09/02-32-07/.hydra/config.yaml b/outputs/2026-06-09/02-32-07/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..286a0f5fe115cc7ce1edcf44dcd08d0631c418e5 --- /dev/null +++ b/outputs/2026-06-09/02-32-07/.hydra/config.yaml @@ -0,0 +1,1003 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_sas + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format, for example: place 1 at + row 2 col 3 or 1,2,3 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 10 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 25 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-09/02-32-07/.hydra/hydra.yaml b/outputs/2026-06-09/02-32-07/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..400e9b8e1fc14258ac144dbd8795e479f7ae781a --- /dev/null +++ b/outputs/2026-06-09/02-32-07/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_sas + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube2_withthink_sas,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-09/02-32-07 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-09/02-32-07/.hydra/overrides.yaml b/outputs/2026-06-09/02-32-07/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..feb1671637361bcea9f0dab2e7cb763fd9b5ed25 --- /dev/null +++ b/outputs/2026-06-09/02-32-07/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_sas diff --git a/outputs/2026-06-09/02-32-07/train.log b/outputs/2026-06-09/02-32-07/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-09/05-39-33/.hydra/config.yaml b/outputs/2026-06-09/05-39-33/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f447d3a5dc08ffa9c095c2f4ba87f67430497398 --- /dev/null +++ b/outputs/2026-06-09/05-39-33/.hydra/config.yaml @@ -0,0 +1,1003 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube3_withthink_sas + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format, for example: place 1 at + row 2 col 3 or 1,2,3 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 3 + max_steps: 10 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 25 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-09/05-39-33/train.log b/outputs/2026-06-09/05-39-33/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-09/11-20-05/train.log b/outputs/2026-06-09/11-20-05/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-09/11-58-39/train.log b/outputs/2026-06-09/11-58-39/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-09/13-00-35/.hydra/overrides.yaml b/outputs/2026-06-09/13-00-35/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9a752fd778f03f732ec7009e649b3089b7892429 --- /dev/null +++ b/outputs/2026-06-09/13-00-35/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sas +- custom_envs.rubikscube.env_config.scramble_depth=3 +- trainer.experiment_name=rubikscube3_withthink_sas diff --git a/outputs/2026-06-09/13-00-35/train.log b/outputs/2026-06-09/13-00-35/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-09/16-00-04/.hydra/config.yaml b/outputs/2026-06-09/16-00-04/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..e4bf7279d1a8a6579678b9d1e8f25e692933a2ee --- /dev/null +++ b/outputs/2026-06-09/16-00-04/.hydra/config.yaml @@ -0,0 +1,1003 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 400 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: sudoku_3b_action1_baseline + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format, for example: place 1 at + row 2 col 3 or 1,2,3 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/share/model/Qwen/Qwen2.5-3B-Instruct +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: -1 + batch_adjust_mode: copy + max_turn: 20 + action_sep: '||' + max_actions_per_turn: 1 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - SimpleSudoku + n_groups: + - 8 + val: + env_groups: 32 + group_size: 16 + env_configs: + tags: + - SimpleSudoku + n_groups: + - 32 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-09/16-05-29/.hydra/config.yaml b/outputs/2026-06-09/16-05-29/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..66b79b9af40d3bcef623bac307e6b49f2a9996e7 --- /dev/null +++ b/outputs/2026-06-09/16-05-29/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 400 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: sudoku_3b_action1_abstraction + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/share/model/Qwen/Qwen2.5-3B-Instruct +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: -1 + batch_adjust_mode: copy + max_turn: 20 + action_sep: '||' + max_actions_per_turn: 1 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - SimpleSudoku + n_groups: + - 8 + val: + env_groups: 32 + group_size: 16 + env_configs: + tags: + - SimpleSudoku + n_groups: + - 32 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-09/16-05-29/.hydra/hydra.yaml b/outputs/2026-06-09/16-05-29/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5778884a90c70c184868f6720d7c6a2bd1d245a0 --- /dev/null +++ b/outputs/2026-06-09/16-05-29/.hydra/hydra.yaml @@ -0,0 +1,178 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: + - pkg://verl.trainer.config + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - model_path=/mnt/general/share/model/Qwen/Qwen2.5-3B-Instruct + - trainer.experiment_name=sudoku_3b_action1_abstraction + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,model_path=/mnt/general/share/model/Qwen/Qwen2.5-3B-Instruct,trainer.experiment_name=sudoku_3b_action1_abstraction,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _8_sudoku + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: verl.trainer.config + schema: pkg + provider: hydra.searchpath in main + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-09/16-05-29 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-09/16-05-29/.hydra/overrides.yaml b/outputs/2026-06-09/16-05-29/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..64a0cc7301b91aa19ddd503b3cc0021a4725b6d1 --- /dev/null +++ b/outputs/2026-06-09/16-05-29/.hydra/overrides.yaml @@ -0,0 +1,5 @@ +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- model_path=/mnt/general/share/model/Qwen/Qwen2.5-3B-Instruct +- trainer.experiment_name=sudoku_3b_action1_abstraction diff --git a/outputs/2026-06-09/17-01-43/.hydra/config.yaml b/outputs/2026-06-09/17-01-43/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1c80cbc98866a50c23cdb06d7c10e1efc141a950 --- /dev/null +++ b/outputs/2026-06-09/17-01-43/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 400 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: sokoban_3b_box1_withthink_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: -1 + batch_adjust_mode: copy + max_turn: 20 + action_sep: '||' + max_actions_per_turn: 1 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - CoordSokoban + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - CoordSokoban + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-09/17-01-43/.hydra/overrides.yaml b/outputs/2026-06-09/17-01-43/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..736504fc836b4f46ad7e2f65e2cc0385d00fd47e --- /dev/null +++ b/outputs/2026-06-09/17-01-43/.hydra/overrides.yaml @@ -0,0 +1,5 @@ +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_sa +- trainer.experiment_name=sokoban_3b_box1_withthink_sa diff --git a/outputs/2026-06-09/17-01-43/train.log b/outputs/2026-06-09/17-01-43/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-09/20-37-55/.hydra/hydra.yaml b/outputs/2026-06-09/20-37-55/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..87c85ef28ab9ae91050f89321162207c195781e7 --- /dev/null +++ b/outputs/2026-06-09/20-37-55/.hydra/hydra.yaml @@ -0,0 +1,178 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: + - pkg://verl.trainer.config + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_sas + - trainer.experiment_name=sokoban_3b_box1_withthink_sas + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_sas,trainer.experiment_name=sokoban_3b_box1_withthink_sas,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _2_sokoban + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: verl.trainer.config + schema: pkg + provider: hydra.searchpath in main + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-09/20-37-55 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-09/20-37-55/.hydra/overrides.yaml b/outputs/2026-06-09/20-37-55/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..475f11baa17afb9f83c85a2c15b48e94fc45ec0c --- /dev/null +++ b/outputs/2026-06-09/20-37-55/.hydra/overrides.yaml @@ -0,0 +1,5 @@ +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_sas +- trainer.experiment_name=sokoban_3b_box1_withthink_sas diff --git a/outputs/2026-06-09/20-37-55/train.log b/outputs/2026-06-09/20-37-55/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-12/14-59-00/.hydra/config.yaml b/outputs/2026-06-12/14-59-00/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a4823193790f7f21aef53985d7c3ef103275a2db --- /dev/null +++ b/outputs/2026-06-12/14-59-00/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 25 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-12/14-59-00/.hydra/hydra.yaml b/outputs/2026-06-12/14-59-00/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..97aaa33f2300dcde4cd4f424b3ea1cb8af4fc2f6 --- /dev/null +++ b/outputs/2026-06-12/14-59-00/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-12/14-59-00 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-12/14-59-00/train.log b/outputs/2026-06-12/14-59-00/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-15/10-55-43/.hydra/config.yaml b/outputs/2026-06-15/10-55-43/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a4823193790f7f21aef53985d7c3ef103275a2db --- /dev/null +++ b/outputs/2026-06-15/10-55-43/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 25 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-15/10-55-43/.hydra/hydra.yaml b/outputs/2026-06-15/10-55-43/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6465df55706cd7e4ad265887b839e82d034eb3ca --- /dev/null +++ b/outputs/2026-06-15/10-55-43/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-15/10-55-43 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-15/10-55-43/.hydra/overrides.yaml b/outputs/2026-06-15/10-55-43/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6c8115d8cf840a67eaf1b773a62d2f9e70ebe7a4 --- /dev/null +++ b/outputs/2026-06-15/10-55-43/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa diff --git a/outputs/2026-06-15/10-56-06/.hydra/config.yaml b/outputs/2026-06-15/10-56-06/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f4a70e4a7e80601767e9316c6fb8e82bdea743aa --- /dev/null +++ b/outputs/2026-06-15/10-56-06/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube3_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 3 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 25 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-15/10-56-06/.hydra/hydra.yaml b/outputs/2026-06-15/10-56-06/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a3577493a9a6fc91f5fc29cda7a0ded45b35b3a8 --- /dev/null +++ b/outputs/2026-06-15/10-56-06/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa + - custom_envs.rubikscube.env_config.scramble_depth=3 + - trainer.experiment_name=rubikscube3_withthink_fulltrajc_sa + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=3,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube3_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-15/10-56-06 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-15/10-56-06/.hydra/overrides.yaml b/outputs/2026-06-15/10-56-06/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..87ddae1e99d1e01dcbfe36c56ff781dd2354f676 --- /dev/null +++ b/outputs/2026-06-15/10-56-06/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa +- custom_envs.rubikscube.env_config.scramble_depth=3 +- trainer.experiment_name=rubikscube3_withthink_fulltrajc_sa diff --git a/outputs/2026-06-15/10-56-06/train.log b/outputs/2026-06-15/10-56-06/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-15/17-29-15/.hydra/config.yaml b/outputs/2026-06-15/17-29-15/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..5f64970b69e535d638744fcc500e56b9fffaed17 --- /dev/null +++ b/outputs/2026-06-15/17-29-15/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_fulltracj_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 25 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-15/17-29-15/.hydra/hydra.yaml b/outputs/2026-06-15/17-29-15/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..ca6d0bdbd02fffda86499371f672be43e455f62b --- /dev/null +++ b/outputs/2026-06-15/17-29-15/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_fulltracj_sa + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_fulltracj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-15/17-29-15 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-15/17-29-15/.hydra/overrides.yaml b/outputs/2026-06-15/17-29-15/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9b03123f9520a5c1cdb68fece1a6773355493194 --- /dev/null +++ b/outputs/2026-06-15/17-29-15/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_fulltracj_sa +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa diff --git a/outputs/2026-06-16/13-25-02/.hydra/hydra.yaml b/outputs/2026-06-16/13-25-02/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..136ef2f6b82990daac2625c031070065c24f871a --- /dev/null +++ b/outputs/2026-06-16/13-25-02/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_fulltracj_sa + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_fulltracj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-16/13-25-02 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-16/13-25-02/.hydra/overrides.yaml b/outputs/2026-06-16/13-25-02/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9b03123f9520a5c1cdb68fece1a6773355493194 --- /dev/null +++ b/outputs/2026-06-16/13-25-02/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_sokoban_box2_withthink_fulltracj_sa +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa diff --git a/outputs/2026-06-16/17-30-16/.hydra/config.yaml b/outputs/2026-06-16/17-30-16/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..be4d5f9dd3dbd3c737fe7c1393e9601fbedfc4b3 --- /dev/null +++ b/outputs/2026-06-16/17-30-16/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-16/17-30-16/.hydra/hydra.yaml b/outputs/2026-06-16/17-30-16/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..dcf0d69c962a4366e3c48e03c1e5a0fc1a3329c5 --- /dev/null +++ b/outputs/2026-06-16/17-30-16/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-16/17-30-16 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-16/17-30-16/.hydra/overrides.yaml b/outputs/2026-06-16/17-30-16/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d2b95b90f7fae5e2efd9b6828b0f2d8f708a6698 --- /dev/null +++ b/outputs/2026-06-16/17-30-16/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa diff --git a/outputs/2026-06-16/17-30-16/train.log b/outputs/2026-06-16/17-30-16/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-16/18-42-33/.hydra/config.yaml b/outputs/2026-06-16/18-42-33/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..be4d5f9dd3dbd3c737fe7c1393e9601fbedfc4b3 --- /dev/null +++ b/outputs/2026-06-16/18-42-33/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: -1 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: checkpoints/${trainer.project_name}/${trainer.experiment_name} + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-16/18-42-33/.hydra/hydra.yaml b/outputs/2026-06-16/18-42-33/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..003a1d117cef9572edf7a643cab5fe9040321693 --- /dev/null +++ b/outputs/2026-06-16/18-42-33/.hydra/hydra.yaml @@ -0,0 +1,177 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-16/18-42-33 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-16/18-42-33/.hydra/overrides.yaml b/outputs/2026-06-16/18-42-33/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d2b95b90f7fae5e2efd9b6828b0f2d8f708a6698 --- /dev/null +++ b/outputs/2026-06-16/18-42-33/.hydra/overrides.yaml @@ -0,0 +1,8 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa diff --git a/outputs/2026-06-16/18-42-33/train.log b/outputs/2026-06-16/18-42-33/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-18/11-52-39/.hydra/config.yaml b/outputs/2026-06-18/11-52-39/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2cd5336518a94c408ec580dc56c59072f7bdc87e --- /dev/null +++ b/outputs/2026-06-18/11-52-39/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube1_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-18/11-52-39/.hydra/hydra.yaml b/outputs/2026-06-18/11-52-39/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c17a90cf6973730e63e98763ca60b7b3d71a46bb --- /dev/null +++ b/outputs/2026-06-18/11-52-39/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa + - custom_envs.rubikscube.env_config.scramble_depth=1 + - trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=1,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl,trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-18/11-52-39 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-18/11-52-39/.hydra/overrides.yaml b/outputs/2026-06-18/11-52-39/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..30784c9f77368a2f9fea7efff56699ce6eea94fd --- /dev/null +++ b/outputs/2026-06-18/11-52-39/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa +- custom_envs.rubikscube.env_config.scramble_depth=1 +- trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl diff --git a/outputs/2026-06-18/11-52-39/train.log b/outputs/2026-06-18/11-52-39/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-18/12-24-12/.hydra/config.yaml b/outputs/2026-06-18/12-24-12/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d88f3bd9f5654df97b8599bf7b0295db12ac86b2 --- /dev/null +++ b/outputs/2026-06-18/12-24-12/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-18/12-24-12/.hydra/hydra.yaml b/outputs/2026-06-18/12-24-12/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..93ce84bb5ba75e67e9eb1b52f4f2c1fd958eccbc --- /dev/null +++ b/outputs/2026-06-18/12-24-12/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa_rl,trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-18/12-24-12 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-18/12-24-12/.hydra/overrides.yaml b/outputs/2026-06-18/12-24-12/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..c2c1cad2ad53076197ce73549abf31438318d0b9 --- /dev/null +++ b/outputs/2026-06-18/12-24-12/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sas +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_fulltrajc_sa +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_fulltraj_sa_rl diff --git a/outputs/2026-06-18/12-24-12/train.log b/outputs/2026-06-18/12-24-12/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-18/13-51-23/.hydra/hydra.yaml b/outputs/2026-06-18/13-51-23/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..737864d57e283ecc9da06a0d635e0b0d17fc5d84 --- /dev/null +++ b/outputs/2026-06-18/13-51-23/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa + - custom_envs.rubikscube.env_config.scramble_depth=3 + - trainer.experiment_name=rubikscube3_withthink_fulltrajc_sa + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=3,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa_rl,trainer.experiment_name=rubikscube3_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-18/13-51-23 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-18/13-51-23/.hydra/overrides.yaml b/outputs/2026-06-18/13-51-23/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8ea6690aacfce3355ce938962ff9b78a5c2a3419 --- /dev/null +++ b/outputs/2026-06-18/13-51-23/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa +- custom_envs.rubikscube.env_config.scramble_depth=3 +- trainer.experiment_name=rubikscube3_withthink_fulltrajc_sa +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_fulltraj_sa_rl diff --git a/outputs/2026-06-18/13-51-23/train.log b/outputs/2026-06-18/13-51-23/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-22/22-00-14/.hydra/config.yaml b/outputs/2026-06-22/22-00-14/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..9fe9522c4cff5679bd3ca8cac8d78dafcf2ca5c4 --- /dev/null +++ b/outputs/2026-06-22/22-00-14/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube1_withthink_sas + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sas_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-22/22-00-14/.hydra/hydra.yaml b/outputs/2026-06-22/22-00-14/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..b0b8890b80cbc4c0c4208df6393793f88df81a8d --- /dev/null +++ b/outputs/2026-06-22/22-00-14/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sas + - custom_envs.rubikscube.env_config.scramble_depth=1 + - trainer.experiment_name=rubikscube1_withthink_sas + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sas_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=1,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sas_rl,trainer.experiment_name=rubikscube1_withthink_sas,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-22/22-00-14 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-22/22-00-14/.hydra/overrides.yaml b/outputs/2026-06-22/22-00-14/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..91be8468d30a8a4d6471c3ce277e4162d8f564d8 --- /dev/null +++ b/outputs/2026-06-22/22-00-14/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sas +- custom_envs.rubikscube.env_config.scramble_depth=1 +- trainer.experiment_name=rubikscube1_withthink_sas +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sas_rl diff --git a/outputs/2026-06-22/22-00-14/train.log b/outputs/2026-06-22/22-00-14/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-22/22-11-12/.hydra/config.yaml b/outputs/2026-06-22/22-11-12/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..6f3c4b3b8c44a40df82c95ace82017be9f8dd906 --- /dev/null +++ b/outputs/2026-06-22/22-11-12/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube1_withthink_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-22/22-11-12/.hydra/hydra.yaml b/outputs/2026-06-22/22-11-12/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d2ebe82d0b6753bef2b56d2038309dcb83da3fef --- /dev/null +++ b/outputs/2026-06-22/22-11-12/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sa + - custom_envs.rubikscube.env_config.scramble_depth=1 + - trainer.experiment_name=rubikscube1_withthink_sa + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=1,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sa_rl,trainer.experiment_name=rubikscube1_withthink_sa,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-22/22-11-12 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-22/22-11-12/.hydra/overrides.yaml b/outputs/2026-06-22/22-11-12/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..a53c86935332d595769e5723191b244ebce11957 --- /dev/null +++ b/outputs/2026-06-22/22-11-12/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sa +- custom_envs.rubikscube.env_config.scramble_depth=1 +- trainer.experiment_name=rubikscube1_withthink_sa +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sa_rl diff --git a/outputs/2026-06-22/22-11-12/train.log b/outputs/2026-06-22/22-11-12/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-22/22-32-28/.hydra/config.yaml b/outputs/2026-06-22/22-32-28/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..311e1d8d67476703485a32dc093412dee4986810 --- /dev/null +++ b/outputs/2026-06-22/22-32-28/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_sas + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sas_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-22/22-32-28/.hydra/hydra.yaml b/outputs/2026-06-22/22-32-28/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d576eb228322b23e2fc1cb4a7e2a5c5abe1c06d2 --- /dev/null +++ b/outputs/2026-06-22/22-32-28/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_sas + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sas_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sas_rl,trainer.experiment_name=rubikscube2_withthink_sas,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-22/22-32-28 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-22/22-32-28/.hydra/overrides.yaml b/outputs/2026-06-22/22-32-28/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..da2eeddf82bc2011a94d49285ab2fb1c6eb0f157 --- /dev/null +++ b/outputs/2026-06-22/22-32-28/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sas +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_sas +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sas_rl diff --git a/outputs/2026-06-22/22-32-28/train.log b/outputs/2026-06-22/22-32-28/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-22/22-48-08/.hydra/config.yaml b/outputs/2026-06-22/22-48-08/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..18d87cd5ea7d863b033c6c85af72058e4f1b5c57 --- /dev/null +++ b/outputs/2026-06-22/22-48-08/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube2_withthink_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 2 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-22/22-48-08/.hydra/hydra.yaml b/outputs/2026-06-22/22-48-08/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d5d5e6f4fb383c96e1859db4a59ff215940de370 --- /dev/null +++ b/outputs/2026-06-22/22-48-08/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sa + - custom_envs.rubikscube.env_config.scramble_depth=2 + - trainer.experiment_name=rubikscube2_withthink_sa + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=2,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sa_rl,trainer.experiment_name=rubikscube2_withthink_sa,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-22/22-48-08 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-22/22-48-08/.hydra/overrides.yaml b/outputs/2026-06-22/22-48-08/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..efa65edd8aadb4c43461bb6f7555edc7bdbe16d5 --- /dev/null +++ b/outputs/2026-06-22/22-48-08/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube2_withthink_sa +- custom_envs.rubikscube.env_config.scramble_depth=2 +- trainer.experiment_name=rubikscube2_withthink_sa +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube2_withthink_sa_rl diff --git a/outputs/2026-06-22/22-48-08/train.log b/outputs/2026-06-22/22-48-08/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-22/23-59-57/.hydra/config.yaml b/outputs/2026-06-22/23-59-57/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..68c05031fdc35ac45a939aa5b97e330978af4c72 --- /dev/null +++ b/outputs/2026-06-22/23-59-57/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube3_withthink_sas + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_sas_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 3 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sas +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-22/23-59-57/.hydra/hydra.yaml b/outputs/2026-06-22/23-59-57/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..024100331023e8f5b2b0fb8a36b41890c6d5a9ee --- /dev/null +++ b/outputs/2026-06-22/23-59-57/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sas + - custom_envs.rubikscube.env_config.scramble_depth=3 + - trainer.experiment_name=rubikscube3_withthink_sas + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_sas_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=3,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sas,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_sas_rl,trainer.experiment_name=rubikscube3_withthink_sas,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-22/23-59-57 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-22/23-59-57/train.log b/outputs/2026-06-22/23-59-57/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-23/00-14-05/.hydra/config.yaml b/outputs/2026-06-23/00-14-05/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..8842672def67ba607884be061b8b0265a2f32c7e --- /dev/null +++ b/outputs/2026-06-23/00-14-05/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 200 + project_name: ragen + experiment_name: rubikscube3_withthink_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 200 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 3 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-23/00-14-05/.hydra/hydra.yaml b/outputs/2026-06-23/00-14-05/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..20d86a6375204f3127524cd2828fd119d47e379a --- /dev/null +++ b/outputs/2026-06-23/00-14-05/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=200 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sa + - custom_envs.rubikscube.env_config.scramble_depth=3 + - trainer.experiment_name=rubikscube3_withthink_sa + - trainer.save_freq=200 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=3,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_sa_rl,trainer.experiment_name=rubikscube3_withthink_sa,trainer.n_gpus_per_node=8,trainer.save_freq=200,trainer.total_training_steps=200 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-23/00-14-05 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-23/00-14-05/.hydra/overrides.yaml b/outputs/2026-06-23/00-14-05/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..3cb961e2364a3e22ea63872f2e34e262f1d95ef6 --- /dev/null +++ b/outputs/2026-06-23/00-14-05/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=200 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube3_withthink_sa +- custom_envs.rubikscube.env_config.scramble_depth=3 +- trainer.experiment_name=rubikscube3_withthink_sa +- trainer.save_freq=200 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube3_withthink_sa_rl diff --git a/outputs/2026-06-23/00-14-05/train.log b/outputs/2026-06-23/00-14-05/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-23/14-24-28/.hydra/config.yaml b/outputs/2026-06-23/14-24-28/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d572a314fc18de2fde27ba413752320de8dc1093 --- /dev/null +++ b/outputs/2026-06-23/14-24-28/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 50 + project_name: ragen + experiment_name: rubikscube1_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 50 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-23/14-24-28/.hydra/hydra.yaml b/outputs/2026-06-23/14-24-28/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..1c69691f7358f69663cae7fc89e3b3ae33434b77 --- /dev/null +++ b/outputs/2026-06-23/14-24-28/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=50 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa + - custom_envs.rubikscube.env_config.scramble_depth=1 + - trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa + - trainer.save_freq=50 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=1,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl,trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.save_freq=50,trainer.total_training_steps=50 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-23/14-24-28 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-23/14-24-28/.hydra/overrides.yaml b/outputs/2026-06-23/14-24-28/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2ab321463fdd102e95777e56bcccbeec360d4fd1 --- /dev/null +++ b/outputs/2026-06-23/14-24-28/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=50 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa +- custom_envs.rubikscube.env_config.scramble_depth=1 +- trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa +- trainer.save_freq=50 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl diff --git a/outputs/2026-06-23/14-24-28/train.log b/outputs/2026-06-23/14-24-28/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-23/14-31-50/.hydra/config.yaml b/outputs/2026-06-23/14-31-50/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..d572a314fc18de2fde27ba413752320de8dc1093 --- /dev/null +++ b/outputs/2026-06-23/14-31-50/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 50 + project_name: ragen + experiment_name: rubikscube1_withthink_fulltrajc_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 50 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-23/14-31-50/.hydra/hydra.yaml b/outputs/2026-06-23/14-31-50/.hydra/hydra.yaml new file mode 100644 index 0000000000000000000000000000000000000000..358cfc21bb3a856a74670462f7305499817254ff --- /dev/null +++ b/outputs/2026-06-23/14-31-50/.hydra/hydra.yaml @@ -0,0 +1,179 @@ +hydra: + run: + dir: outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} + sweep: + dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} + subdir: ${hydra.job.num} + launcher: + _target_: hydra._internal.core_plugins.basic_launcher.BasicLauncher + sweeper: + _target_: hydra._internal.core_plugins.basic_sweeper.BasicSweeper + max_batch_size: null + params: null + help: + app_name: ${hydra.job.name} + header: '${hydra.help.app_name} is powered by Hydra. + + ' + footer: 'Powered by Hydra (https://hydra.cc) + + Use --hydra-help to view Hydra specific help + + ' + template: '${hydra.help.header} + + == Configuration groups == + + Compose your configuration from those groups (group=option) + + + $APP_CONFIG_GROUPS + + + == Config == + + Override anything in the config (foo.bar=value) + + + $CONFIG + + + ${hydra.help.footer} + + ' + hydra_help: + template: 'Hydra (${hydra.runtime.version}) + + See https://hydra.cc for more info. + + + == Flags == + + $FLAGS_HELP + + + == Configuration groups == + + Compose your configuration from those groups (For example, append hydra/job_logging=disabled + to command line) + + + $HYDRA_CONFIG_GROUPS + + + Use ''--cfg hydra'' to Show the Hydra config. + + ' + hydra_help: ??? + hydra_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][HYDRA] %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + root: + level: INFO + handlers: + - console + loggers: + logging_example: + level: DEBUG + disable_existing_loggers: false + job_logging: + version: 1 + formatters: + simple: + format: '[%(asctime)s][%(name)s][%(levelname)s] - %(message)s' + handlers: + console: + class: logging.StreamHandler + formatter: simple + stream: ext://sys.stdout + file: + class: logging.FileHandler + formatter: simple + filename: ${hydra.runtime.output_dir}/${hydra.job.name}.log + root: + level: INFO + handlers: + - console + - file + disable_existing_loggers: false + env: {} + mode: RUN + searchpath: [] + callbacks: {} + output_subdir: .hydra + overrides: + hydra: + - hydra.mode=RUN + task: + - system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' + - actor_rollout_ref.rollout.rollout_filter_strategy=top_p + - actor_rollout_ref.rollout.rollout_filter_value=0.9 + - trainer.total_training_steps=50 + - trainer.n_gpus_per_node=8 + - model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa + - custom_envs.rubikscube.env_config.scramble_depth=1 + - trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa + - trainer.save_freq=50 + - trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl + job: + name: train + chdir: null + override_dirname: actor_rollout_ref.rollout.rollout_filter_strategy=top_p,actor_rollout_ref.rollout.rollout_filter_value=0.9,custom_envs.rubikscube.env_config.scramble_depth=1,model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa,system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7',trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl,trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa,trainer.n_gpus_per_node=8,trainer.save_freq=50,trainer.total_training_steps=50 + id: ??? + num: ??? + config_name: _10_rubikscube + env_set: {} + env_copy: [] + config: + override_dirname: + kv_sep: '=' + item_sep: ',' + exclude_keys: [] + runtime: + version: 1.3.2 + version_base: '1.3' + cwd: /mnt/general/wanghy/RAGEN_v2 + config_sources: + - path: hydra.conf + schema: pkg + provider: hydra + - path: /mnt/general/wanghy/RAGEN_v2/config + schema: file + provider: main + - path: /mnt/general/wanghy/RAGEN_v2/verl/verl/trainer/config + schema: file + provider: command-line + - path: '' + schema: structured + provider: schema + output_dir: /mnt/general/wanghy/RAGEN_v2/outputs/2026-06-23/14-31-50 + choices: + reward_model: dp_reward_model + critic: dp_critic + critic/../engine@critic.model.fsdp_config: fsdp + critic/../optim@critic.optim: fsdp + model@actor_rollout_ref.model: hf_model + rollout@actor_rollout_ref.rollout: rollout + ref@actor_rollout_ref.ref: dp_ref + ref/../engine@actor_rollout_ref.ref.fsdp_config: fsdp + data: legacy_data + actor@actor_rollout_ref.actor: dp_actor + actor/../engine@actor_rollout_ref.actor.fsdp_config: fsdp + actor/../optim@actor_rollout_ref.actor.optim: fsdp + hydra/env: default + hydra/callbacks: null + hydra/job_logging: default + hydra/hydra_logging: default + hydra/hydra_help: default + hydra/help: default + hydra/sweeper: basic + hydra/launcher: basic + hydra/output: default + verbose: false diff --git a/outputs/2026-06-23/14-31-50/.hydra/overrides.yaml b/outputs/2026-06-23/14-31-50/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..2ab321463fdd102e95777e56bcccbeec360d4fd1 --- /dev/null +++ b/outputs/2026-06-23/14-31-50/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=50 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa +- custom_envs.rubikscube.env_config.scramble_depth=1 +- trainer.experiment_name=rubikscube1_withthink_fulltrajc_sa +- trainer.save_freq=50 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_fulltraj_sa_rl diff --git a/outputs/2026-06-23/14-31-50/train.log b/outputs/2026-06-23/14-31-50/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-23/14-51-00/.hydra/config.yaml b/outputs/2026-06-23/14-51-00/.hydra/config.yaml new file mode 100644 index 0000000000000000000000000000000000000000..f4f782250d818c9cb21788f050e92589f2133fa0 --- /dev/null +++ b/outputs/2026-06-23/14-51-00/.hydra/config.yaml @@ -0,0 +1,1048 @@ +actor_rollout_ref: + actor: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-06 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + _target_: verl.workers.config.FSDPActorConfig + strategy: fsdp + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: false + ppo_max_token_len_per_gpu: 16384 + clip_ratio: 0.2 + clip_ratio_low: 0.2 + clip_ratio_high: 0.28 + freeze_vision_tower: false + policy_loss: + _target_: verl.workers.config.PolicyLossConfig + loss_mode: vanilla + clip_cov_ratio: 0.0002 + clip_cov_lb: 1.0 + clip_cov_ub: 5.0 + kl_cov_ratio: 0.0002 + ppo_kl_coef: 0.1 + clip_ratio_c: 3.0 + loss_agg_mode: token-mean + entropy_coeff: 0.001 + tis_imp_ratio_cap: -1 + use_kl_loss: false + use_torch_compile: true + kl_loss_coef: 0.0 + kl_loss_type: kl + ppo_epochs: 1 + shuffle: false + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + use_fused_kernels: ${oc.select:actor_rollout_ref.model.use_fused_kernels,false} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + grad_clip: 1.0 + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + use_remove_padding: ${oc.select:actor_rollout_ref.model.use_remove_padding,false} + use_ref: true + grpo_advantage_length_weight: ${grpo_advantage_length_weight} + filter_loss_scaling: none + ref: + strategy: ${actor_rollout_ref.actor.strategy} + use_torch_compile: ${oc.select:actor_rollout_ref.actor.use_torch_compile,true} + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + model: null + ulysses_sequence_parallel_size: ${oc.select:actor_rollout_ref.actor.ulysses_sequence_parallel_size,1} + entropy_from_logits_with_chunking: false + entropy_checkpointing: false + rollout: + _target_: verl.workers.config.RolloutConfig + name: vllm + mode: sync + temperature: 1 + top_k: -1 + top_p: 1 + prompt_length: 1 + response_length: 500 + dtype: bfloat16 + gpu_memory_utilization: 0.7 + ignore_eos: false + enforce_eager: true + cudagraph_capture_sizes: null + free_cache_engine: true + tensor_model_parallel_size: 1 + data_parallel_size: 1 + expert_parallel_size: 1 + max_num_batched_tokens: 16384 + max_model_len: 16384 + max_num_seqs: 1024 + enable_chunked_prefill: true + enable_prefix_caching: true + load_format: auto + log_prob_micro_batch_size: null + log_prob_micro_batch_size_per_gpu: ${log_prob_micro_batch_size_per_gpu} + log_prob_use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + log_prob_max_token_len_per_gpu: ${oc.select:actor_rollout_ref.actor.ppo_max_token_len_per_gpu,16384} + disable_log_stats: true + do_sample: true + 'n': 1 + over_sample_rate: 0 + multi_stage_wake_up: false + engine_kwargs: + vllm: {} + sglang: {} + val_kwargs: + _target_: verl.workers.config.SamplingConfig + top_k: -1 + top_p: 1.0 + temperature: 0.5 + 'n': 1 + do_sample: true + multi_turn: + _target_: verl.workers.config.MultiTurnConfig + enable: false + max_assistant_turns: null + tool_config_path: null + max_user_turns: null + max_parallel_calls: 1 + max_tool_response_length: 256 + tool_response_truncate_side: middle + interaction_config_path: null + use_inference_chat_template: false + tokenization_sanity_check_mode: strict + format: hermes + num_repeat_rollouts: null + calculate_log_probs: false + agent: + _target_: verl.workers.config.AgentLoopConfig + num_workers: 8 + agent_loop_config_path: null + custom_async_server: + _target_: verl.workers.config.CustomAsyncServerConfig + path: null + name: null + update_weights_bucket_megabytes: 512 + trace: + _target_: verl.workers.config.TraceConfig + backend: null + token2text: false + skip_rollout: false + skip_dump_dir: /tmp/rollout_dump + skip_tokenizer_init: true + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: ${oc.select:actor_rollout_ref.actor.profiler.enable,false} + all_ranks: ${oc.select:actor_rollout_ref.actor.profiler.all_ranks,false} + ranks: ${oc.select:actor_rollout_ref.actor.profiler.ranks,[]} + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + layered_summon: false + rollout_filter_value: 0.9 + rollout_filter_strategy: top_p + rollout_filter_type: largest + rollout_filter_include_zero: true + rollout_filter_top_p_prob_mode: linear + rollout_filter_selection_eps: 0.01 + rollout_filter_empty_stop_steps: 5 + rollout_filter_metric: reward_variance + gradient_analysis_num_buckets: 6 + gradient_analysis_bucket_mode: quantile + model: + _target_: verl.workers.config.HFModelConfig + path: ${model_path} + hf_config_path: null + tokenizer_path: null + use_shm: false + trust_remote_code: false + custom_chat_template: null + external_lib: null + override_config: {} + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + exclude_modules: null + use_liger: false + use_fused_kernels: false + fused_kernel_options: + impl_backend: torch + hybrid_engine: true + nccl_timeout: 600 +data: + tokenizer: null + use_shm: false + train_files: ~/data/rlhf/gsm8k/train.parquet + val_files: ~/data/rlhf/gsm8k/test.parquet + prompt_key: prompt + reward_fn_key: data_source + max_prompt_length: null + max_response_length: null + train_batch_size: null + val_batch_size: null + return_raw_input_ids: false + return_raw_chat: false + return_full_prompt: false + shuffle: true + dataloader_num_workers: 8 + validation_shuffle: false + filter_overlong_prompts: false + filter_overlong_prompts_workers: 1 + truncation: error + image_key: images + video_key: videos + trust_remote_code: false + custom_cls: + path: null + name: null + return_multi_modal_inputs: true + sampler: + class_path: null + class_name: null + datagen: + path: null + name: null + apply_chat_template_kwargs: {} +critic: + optim: + _target_: verl.workers.config.FSDPOptimizerConfig + lr: 1.0e-05 + lr_warmup_steps_ratio: 0.0 + total_training_steps: -1 + weight_decay: 0.01 + lr_warmup_steps: -1 + betas: + - 0.9 + - 0.999 + clip_grad: 1.0 + min_lr_ratio: 0.0 + num_cycles: 0.5 + warmup_style: constant + model: + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + optimizer_offload: false + offload_policy: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + model_dtype: fp32 + use_orig_params: false + ulysses_sequence_parallel_size: 1 + entropy_from_logits_with_chunking: false + use_torch_compile: true + entropy_checkpointing: false + forward_only: false + strategy: fsdp + path: ${model_path} + tokenizer_path: ${oc.select:actor_rollout_ref.model.path,"~/models/deepseek-llm-7b-chat"} + override_config: {} + external_lib: ${oc.select:actor_rollout_ref.model.external_lib,null} + trust_remote_code: ${oc.select:actor_rollout_ref.model.trust_remote_code,false} + _target_: verl.workers.config.FSDPCriticModelCfg + use_shm: false + enable_gradient_checkpointing: true + enable_activation_offload: false + use_remove_padding: false + lora_rank: ${lora.rank} + lora_alpha: ${lora.alpha} + target_modules: ${lora.target_modules} + _target_: verl.workers.config.FSDPCriticConfig + rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1} + strategy: fsdp + enable: null + ppo_mini_batch_size: ${ppo_mini_batch_size} + ppo_micro_batch_size: null + ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} + use_dynamic_bsz: ${oc.select:actor_rollout_ref.actor.use_dynamic_bsz,false} + ppo_max_token_len_per_gpu: 32768 + forward_max_token_len_per_gpu: ${.ppo_max_token_len_per_gpu} + ppo_epochs: ${oc.select:actor_rollout_ref.actor.ppo_epochs,1} + shuffle: ${oc.select:actor_rollout_ref.actor.shuffle,false} + cliprange_value: 0.5 + loss_agg_mode: ${oc.select:actor_rollout_ref.actor.loss_agg_mode,token-mean} + checkpoint: + _target_: verl.trainer.config.CheckpointConfig + save_contents: + - model + - optimizer + - extra + load_contents: ${.save_contents} + async_save: false + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: ${oc.select:global_profiler.global_tool_config.nsys.discrete} + npu: + _target_: verl.utils.profiler.config.NPUToolConfig + contents: [] + level: level1 + analysis: true + discrete: false + torch: + _target_: verl.utils.profiler.config.TorchProfilerToolConfig + step_start: 0 + step_end: null + torch_memory: + _target_: verl.utils.profiler.config.TorchMemoryToolConfig + trace_alloc_max_entries: ${oc.select:global_profiler.global_tool_config.torch_memory.trace_alloc_max_entries,100000} + stack_depth: ${oc.select:global_profiler.global_tool_config.torch_memory.stack_depth,32} + forward_micro_batch_size: ${oc.select:.ppo_micro_batch_size,null} + forward_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size_per_gpu,null} + ulysses_sequence_parallel_size: 1 + grad_clip: 1.0 +reward_model: + enable: false + enable_resource_pool: false + n_gpus_per_node: 0 + nnodes: 0 + strategy: fsdp + model: + input_tokenizer: ${actor_rollout_ref.model.path} + path: ~/models/FsfairX-LLaMA3-RM-v0.1 + external_lib: ${actor_rollout_ref.model.external_lib} + trust_remote_code: false + use_shm: false + use_remove_padding: false + use_fused_kernels: ${actor_rollout_ref.model.use_fused_kernels} + fsdp_config: + _target_: verl.workers.config.FSDPEngineConfig + wrap_policy: + min_num_params: 0 + param_offload: false + reshard_after_forward: true + fsdp_size: -1 + forward_prefetch: false + micro_batch_size: null + micro_batch_size_per_gpu: null + max_length: null + use_dynamic_bsz: ${critic.use_dynamic_bsz} + forward_max_token_len_per_gpu: ${critic.forward_max_token_len_per_gpu} + reward_manager: naive + launch_reward_fn_async: false + sandbox_fusion: + url: null + max_concurrent: 64 + memory_limit_mb: 1024 + profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: ${oc.select:global_profiler.tool,null} + enable: false + all_ranks: false + ranks: [] + save_path: ${oc.select:global_profiler.save_path,null} + tool_config: ${oc.select:actor_rollout_ref.actor.profiler.tool_config,null} + ulysses_sequence_parallel_size: 1 +custom_reward_function: + path: null + name: compute_score +algorithm: + _target_: verl.trainer.config.AlgoConfig + gamma: 1.0 + lam: 1.0 + adv_estimator: gae + norm_adv_by_std_in_grpo: true + use_kl_in_reward: false + kl_penalty: kl + kl_ctrl: + _target_: verl.trainer.config.KLControlConfig + type: fixed + kl_coef: 0.0 + horizon: 10000 + target_kl: 0.1 + use_pf_ppo: false + pf_ppo: + reweight_method: pow + weight_pow: 2.0 + high_level_gamma: 0.95 + bi_level_gae: false + zero_task_advantage: false + soft_advantage_reweight: false +trainer: + balance_batch: true + total_epochs: 30 + total_training_steps: 50 + project_name: ragen + experiment_name: rubikscube1_withthink_sa + logger: + - console + - wandb + log_val_generations: 0 + rollout_data_dir: null + validation_data_dir: null + nnodes: 1 + n_gpus_per_node: 8 + save_freq: 50 + esi_redundant_time: 0 + resume_mode: auto + resume_from_path: null + val_before_train: true + val_only: false + test_freq: 10 + critic_warmup: 0 + default_hdfs_dir: null + del_local_ckpt_after_load: false + default_local_dir: /mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sa_rl + max_actor_ckpt_to_keep: 1 + max_critic_ckpt_to_keep: 1 + ray_wait_register_center_timeout: 300 + device: cuda + use_legacy_worker_impl: auto + local_log_dir: results/ + validation_steps: 1 + generations_to_log_to_wandb: + val: 20 + log_group_rv_table: false + gradient_analysis_mode: false + gradient_analysis_every: 50 + gradient_analysis_env_groups: null + gradient_analysis_group_size: null + gradient_analysis_log_prefilter: false + gradient_analysis_only: false + exit_after_gradient_analysis: false +global_profiler: + _target_: verl.utils.profiler.ProfilerConfig + tool: null + steps: null + profile_continuous_steps: false + save_path: outputs/profile + global_tool_config: + nsys: + _target_: verl.utils.profiler.config.NsightToolConfig + discrete: false + controller_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + worker_nsight_options: + trace: cuda,nvtx,cublas,ucx + cuda-memory-usage: 'true' + cuda-graph-trace: graph + capture-range: cudaProfilerApi + capture-range-end: null + kill: none + torch_memory: + trace_alloc_max_entries: 100000 + stack_depth: 32 + context: all + stacks: all + kw_args: {} +ray_kwargs: + ray_init: + num_cpus: null + timeline_json_file: null +custom_envs: + SimpleSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are a careful Sokoban solver. Use these reusable abstractions\ + \ when reasoning:\n1. First compare the box position with the target position;\ + \ the useful push directions are usually the directions that reduce their row/column\ + \ distance.\n2. Before pushing, move the player to the square opposite the intended\ + \ push direction. A move that only repositions the player can be useful if it\ + \ sets up the next push.\n3. Never push a box into a wall, corner, or narrow\ + \ dead end unless that square is the target or clearly on the only path to the\ + \ target.\n4. Prefer short plans that move the single box steadily toward the\ + \ target; avoid wandering moves that do not improve player position or box position.\n\ + 5. If the box and target are aligned in the same row or column, try to keep\ + \ the box on that line and push along it.\n6. If they are not aligned, first\ + \ push to fix one coordinate, then reposition and push to fix the other coordinate.\n\ + 7. Check that after each push, the player can still reach the next required\ + \ pushing side of the box.\n8. Output only the required XML-like format and\ + \ choose valid Sokoban actions.\n\nYou are the player and you need to push all\ + \ boxes to targets. \nWhen you are right next to a box, you can push it by moving\ + \ in the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box. \nThe answer should be a sequence of actions, like Right\ + \ || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 2 + max_steps: 100 + LargerSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 8 + dim_y: 8 + num_boxes: 2 + max_steps: 100 + search_depth: 10 + SokobanDifferentGridVocab: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: "You are solving the Sokoban puzzle. \nYou are the player and\ + \ you need to push all boxes to targets. \nWhen you are right next to a box,\ + \ you can push it by moving in the same direction. \nYou cannot push a box through\ + \ a wall, and you cannot pull a box. \nThe answer should be a sequence of actions,\ + \ like Right || Right || Up\n" + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + search_depth: 30 + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + grid_lookup: + 0: W + 1: . + 2: G + 3: C + 4: B + 5: A + 6: '@' + grid_vocab: + W: wall + .: empty + G: target + C: box on target + B: box + A: player + '@': player on target + CoordSokoban: + env_type: sokoban + max_actions_per_traj: 20 + env_instruction: "You are solving the Sokoban puzzle. You are the player and you\ + \ need to push all boxes to targets.\nYou are provided with a symbol grid and\ + \ the zero-indexed coordinates of the player, each box, and each target. \n\ + Coordinates range from the top-left corner (0, 0) to the bottom-right corner\ + \ (5, 5). \nWhen you are exactly next to a box, you can push it by moving in\ + \ the same direction. \nYou cannot push a box through a wall, and you cannot\ + \ pull a box.\nThe answer should be a sequence of actions, like Right\ + \ || Right || Up.\n" + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + observation_format: grid_coord + VisualSimpleSokoban: + env_type: sokoban + max_actions_per_traj: 10 + env_instruction: You are solving the Sokoban puzzle. You are the player and you + need to push all boxes to targets. When you are right next to a box, you can + push it by moving in the same direction. You cannot push a box through a wall, + and you cannot pull a box. The answer should be a sequence of actions, like + Right || Right || Up + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + dim_x: 6 + dim_y: 6 + num_boxes: 1 + max_steps: 100 + render_mode: rgb_array + Alfworld: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_in_distribution + AlfworldOOD: + env_type: alfworld + max_actions_per_traj: 50 + parallel_friendly: false + max_workers: 1 + env_instruction: 'You are an expert agent in the ALFRED Embodied Environment. + + Complete household tasks by navigating and interacting with objects. + + + You should first reason step-by-step about the current situation. This reasoning + process MUST be enclosed within tags. + + Once you''ve finished your reasoning, you should choose an admissible action + for current step and present it within ... tags. + + ' + max_tokens: 512 + env_config: + eval_dataset: eval_out_of_distribution + Countdown: + env_type: countdown + max_actions_per_traj: 1 + env_instruction: 'You are solving the Countdown puzzle. You should use the num + list to create an equation that equals the target. Example answer format: + To find an equation using [3, 5, 2] to get 4. Let''s check 2 + 5 = 7, 7 - 3 + = 4. So the answer is 2 + 5 - 3 = 4. 2 + 5 - 3' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: null + Bandit: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: train + BanditTest: + env_type: bandit + max_actions_per_traj: 1 + env_instruction: '' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + split: test + DeepCoder: + env_type: deepcoder + max_actions_per_traj: 1 + env_instruction: 'You are solving a coding task. Provide a complete Python function + solution only. Format: ...' + max_tokens: 8000 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 1 + FrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. Forbid the whole and + go to the target. You may move to the unintended direction due to the slippery + ice. Example answer format: To forbid the hole and go to the target, + I should go left then go up.Left || Up' + max_tokens: 100 + parallel_friendly: false + max_workers: 32 + env_config: + success_rate: 0.8 + CoordFrozenLake: + env_type: frozen_lake + max_actions_per_traj: 10 + env_instruction: 'You are solving the FrozenLake puzzle. The observation includes + both a symbol grid and zero-indexed coordinates for the start, goal, player, + and any holes. + + Coordinates range from the top-left corner (0, 0) to the bottom-right corner + (5, 5). + + Beware that the ice is slippery, so the agent might slide and end up in an unintended + tile. + + Respond with a sequence of actions such as Left || Up || Up. + + ' + max_tokens: 120 + parallel_friendly: false + max_workers: 32 + env_config: + observation_format: grid_coord + success_rate: 0.8 + MetamathQA: + env_type: metamathqa + max_actions_per_traj: 1 + env_instruction: 'You are solving Math problems. ' + max_tokens: 100 + env_config: null + WebShopFull: + env_type: webshop + max_actions_per_traj: 15 + env_instruction: You are an expert autonomous agent operating in the WebShop e‑commerce + environment. + max_tokens: 200 + env_config: + dataset: full + WebShop: + env_type: webshop + max_actions_per_traj: 9 + env_instruction: 'You are browsing an online shop. Based on the instruction, buy + a product that close to the production description. You need to search, read + the search results, pick a product, choose the size and color and buy. You should + only choose action from the available actions list provided later. Example + process: I need a gingko light and 20x20 pillow cover that is hand painted. + First search[gingko light 20x20 pillow cover hand painted], answer format: search[blanket + with fleece throw]. Valid answer is search[] or click[].' + max_tokens: 200 + env_config: + dataset: small + Lean: + env_type: lean + max_actions_per_traj: 30 + env_instruction: You are a Lean theorem prover. Given a Lean theorem statement, + propose a sequence of tactics that completes the proof. Think step by step about + which tactics to apply next. Provide tactics separated by '||', for example + intro || simp || rfl. + max_tokens: 512 + parallel_friendly: true + max_workers: 32 + env_config: null + SimpleSudoku: + env_type: sudoku + max_actions_per_traj: 20 + env_instruction: 'You are a careful 4x4 Sudoku solver. Use these reusable abstractions + when reasoning: + + 1. In , explain the constraint logic behind the move, not just the final + placement. Mention which row, column, or 2x2 box makes the move safe or forced. + + 2. Treat each move as a constraint-preserving placement: place a number only + if it does not already appear in the same row, column, or 2x2 box. + + 3. First look for highly forced cells. If an empty cell has only one legal candidate + after checking its row, column, and box, fill it immediately and explain why + other numbers are excluded. + + 4. If no cell is obviously forced, scan each row, column, and 2x2 box for missing + numbers. If a missing number can go in only one empty position within that unit, + place it there and state that unit-level reason. + + 5. Prefer moves that reduce uncertainty and create new forced cells for the + next turn. A good move should make the remaining puzzle more constrained, not + more ambiguous. + + 6. Solve incrementally: choose exactly one placement, explain it briefly in + , then output that one move in . + + 7. Never modify bracketed initial cells or already-filled cells. Only place + numbers into dots. + + 8. Avoid exploratory guesses when a forced move exists. In this environment, + reliable progress usually comes from constraint propagation rather than trial-and-error. + + 9. If a previous move was invalid or the board did not change, do not repeat + the same placement. Recompute legal candidates and explain the corrected constraint-consistent + choice. + + 10. Use row, column, and box agreement as confidence: the strongest placements + are those supported by multiple constraints at once. + + 11. Keep concise but meaningful: identify the target cell, list or compare + its legal candidates, and give the decisive constraint. + + 12. Output only the required XML-like format and choose valid Sudoku placements. + + + You are solving a Sudoku puzzle. Fill in the grid so that every row, column, + and 2x2 box contains the numbers 1-4 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 1 at row 2 col 3 + + Always output: [brief constraint-based reasoning] + [one placement] + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 4 + difficulty: easy + render_format: with_feedback + show_conflicts: false + show_valid_numbers: false + max_steps: 20 + MediumSudoku: + env_type: sudoku + max_actions_per_traj: 30 + env_instruction: 'You are solving a Sudoku puzzle. Fill in the grid so that every + row, column, and 3x3 box contains the numbers 1-9 without repetition. + + Initial cells are shown in [brackets] and cannot be modified. Empty cells are + shown as dots (.). + + Place numbers one at a time using the format: place 5 at row 2 col 3 + or 2,3,5 + + The environment will provide feedback on valid/invalid moves and show conflicts + if any occur. + + ' + max_tokens: 150 + parallel_friendly: false + max_workers: 32 + env_config: + grid_size: 9 + difficulty: medium + render_format: with_feedback + show_conflicts: true + show_valid_numbers: true + max_steps: 81 + SearchQA: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + SearchQAMock: + env_type: search + max_actions_per_traj: 10 + env_instruction: "You are a search agent answering questions by searching for\ + \ information.\nUse search[your query] to find relevant documents, and finish[your\ + \ answer] to submit your final answer.\n\nYou should first reason step-by-step\ + \ about the current situation. This reasoning process MUST be enclosed within\ + \ tags.\nThen provide your action within ...\ + \ tags.\n\nExamples:\n I need to find information about Ben Platt's\ + \ father.search[Ben Platt father parent]\n Based\ + \ on the search results, Ben Platt's father is Henry Platt.finish[Henry\ + \ Platt]\n" + max_tokens: 300 + parallel_friendly: true + max_workers: 32 + env_config: + max_steps: 10 + max_search_results: 5 + mock_mode: true + game_2048: + env_type: game_2048 + max_actions_per_traj: 700 + env_instruction: 'You are playing the 2048 game on a 4x4 grid. Merge equal tiles + by sliding Up, Right, Down, or Left. + + If a move is invalid (no tiles move), a small penalty is applied. Respond with + a single action. + + Example: Up + + ' + max_tokens: 8192 + env_config: null + rubikscube: + env_type: rubikscube + max_actions_per_traj: 10 + env_instruction: 'You are solving a 2x2 Rubik''s Cube (Pocket Cube). The goal + is to restore the cube so that each of the faces consists of a single, unique + color. + + Available actions use standard Singmaster notation for face rotations: U, U'', + D, D'', L, L'', R, R'', F, F'', B, B''. + + - Faces: U (Up), D (Down), L (Left), R (Right), F (Front), B (Back). + + - Modifiers: A letter alone means 90° clockwise (e.g., ''R''). A letter with + prime ('') means 90° counter-clockwise (e.g., "R''"). + + Respond with a sequence of actions separated by "||". + + Example: U + + ' + max_tokens: 96 + env_config: + scramble_depth: 1 + max_steps: 7 + render_mode: text +system: + CUDA_VISIBLE_DEVICES: 0,1,2,3,4,5,6,7 +seed: + train: 10000 + val: 123 +micro_batch_size_per_gpu: 1 +log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} +ppo_mini_batch_size: 32 +model_path: /mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sa +enable_response_mask: true +grpo_advantage_length_weight: false +lora: + rank: 0 + alpha: 64 + target_modules: all-linear +agent_proxy: + context_window_mode: full + max_context_window: 5 + batch_adjust_mode: copy + max_turn: 10 + action_sep: '||' + max_actions_per_turn: 4 + use_turn_scores: false + enable_think: true + reward_normalization: + grouping: state + method: identity +collapse_detection: + compute_freq: 5 + micro_batch_size: 128 + first_turn_enabled: true + multi_turn_enabled: true + num_samples: 64 +es_manager: + format_penalty: -0.1 + train: + env_groups: 8 + group_size: 16 + env_configs: + tags: + - rubikscube + n_groups: + - 8 + val: + env_groups: 512 + group_size: 1 + env_configs: + tags: + - rubikscube + n_groups: + - 512 +ctx_manager: + generation: + gen_config: + response_length: ${actor_rollout_ref.rollout.response_length} + temperature: ${actor_rollout_ref.rollout.temperature} + top_p: ${actor_rollout_ref.rollout.top_p} + top_k: ${actor_rollout_ref.rollout.top_k} + kwargs: null diff --git a/outputs/2026-06-23/14-51-00/.hydra/overrides.yaml b/outputs/2026-06-23/14-51-00/.hydra/overrides.yaml new file mode 100644 index 0000000000000000000000000000000000000000..780ec0aedad0f87e4e784d15030303637dfd33ef --- /dev/null +++ b/outputs/2026-06-23/14-51-00/.hydra/overrides.yaml @@ -0,0 +1,10 @@ +- system.CUDA_VISIBLE_DEVICES='0,1,2,3,4,5,6,7' +- actor_rollout_ref.rollout.rollout_filter_strategy=top_p +- actor_rollout_ref.rollout.rollout_filter_value=0.9 +- trainer.total_training_steps=50 +- trainer.n_gpus_per_node=8 +- model_path=/mnt/general/wanghy/LLaMA-Factory/saves/qwen3/full/sft/qwen2.5_3B_it_rubikscube1_withthink_sa +- custom_envs.rubikscube.env_config.scramble_depth=1 +- trainer.experiment_name=rubikscube1_withthink_sa +- trainer.save_freq=50 +- trainer.default_local_dir=/mnt/general/wanghy/RAGEN_v2/saves/qwen2.5_3B_it_rubikscube1_withthink_sa_rl diff --git a/outputs/2026-06-23/14-51-00/train.log b/outputs/2026-06-23/14-51-00/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/outputs/2026-06-23/15-19-48/train.log b/outputs/2026-06-23/15-19-48/train.log new file mode 100644 index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391 diff --git a/patches/verl_checkpoint_resharding/README.md b/patches/verl_checkpoint_resharding/README.md new file mode 100644 index 0000000000000000000000000000000000000000..12fef4842c5e5ca31e48a0f83743319746db9362 --- /dev/null +++ b/patches/verl_checkpoint_resharding/README.md @@ -0,0 +1,35 @@ +# FSDP Checkpoint Resharding Patch + +## Problem + +verl's `FSDPCheckpointManager` saves checkpoints as `model_world_size_{N}_rank_{R}.pt`. When resuming with a **different number of GPUs** than the original training run, the checkpoint files don't match and loading fails: + +``` +FileNotFoundError: model_world_size_4_rank_1.pt +# (checkpoint was saved with 1 GPU as model_world_size_1_rank_0.pt) +``` + +## Fix + +This patched `fsdp_checkpoint_manager.py` adds automatic resharding: when the expected checkpoint file doesn't exist but a `model_world_size_1_rank_0.pt` is found, it loads the full state dict and lets FSDP reshard across the current world size. + +### What changes: + +1. **Model loading**: Detects world_size mismatch, switches from `SHARDED_STATE_DICT` to `FULL_STATE_DICT` loading mode so FSDP can reshard automatically +2. **Optimizer**: Skipped during resharding (optimizer state is world_size-specific and must reinitialize) +3. **Extra state** (lr_scheduler, rng): Falls back to the 1-GPU file if available, or skips gracefully + +## How to Apply + +Copy the patched file over the verl submodule file: + +```bash +cp patches/verl_checkpoint_resharding/fsdp_checkpoint_manager.py \ + verl/verl/utils/checkpoint/fsdp_checkpoint_manager.py +``` + +## Limitations + +- Only supports resharding **from 1 GPU** to N GPUs (not arbitrary M→N) +- Optimizer state is not transferred (will reinitialize from scratch) +- LR scheduler state may not transfer if extra state was not saved diff --git a/ragen/__init__.py b/ragen/__init__.py new file mode 100644 index 0000000000000000000000000000000000000000..e041332d026eb0b7bf64753cc1cdd2b5c5c225a1 --- /dev/null +++ b/ragen/__init__.py @@ -0,0 +1,6 @@ +"""RAGEN package initialisation.""" + +from ragen.patches import apply_omega_conf_patch + +# Ensure VERL config instantiation accepts RAGEN-specific extensions. +apply_omega_conf_patch() diff --git a/ragen/__pycache__/__init__.cpython-310.pyc b/ragen/__pycache__/__init__.cpython-310.pyc new file mode 100644 index 0000000000000000000000000000000000000000..30bb098a78dfe35cb2c4b48ddcd06d55501a1e91 Binary files /dev/null and b/ragen/__pycache__/__init__.cpython-310.pyc differ diff --git a/ragen/env/__pycache__/base.cpython-310.pyc b/ragen/env/__pycache__/base.cpython-310.pyc new file mode 100644 index 0000000000000000000000000000000000000000..d2dab25afd12e228be452ed677ea40845b550d6c Binary files /dev/null and b/ragen/env/__pycache__/base.cpython-310.pyc differ diff --git a/ragen/env/alfworld/env.py b/ragen/env/alfworld/env.py new file mode 100644 index 0000000000000000000000000000000000000000..81af4e10f669153294788228e29c258c3d930584 --- /dev/null +++ b/ragen/env/alfworld/env.py @@ -0,0 +1,248 @@ +""" +This is the environment for the ALFRED dataset. +author: Qineng Wang +date: 2025-03-30 + +Modified to match verl-agent AlfWorld implementation: +- Dynamic mode switching between train/val +- Simplified reward: score * won +- Admissible actions included in observation +""" +import os +import random +import textworld +import textworld.gym +import numpy as np +from alfworld.agents.environment.alfred_tw_env import AlfredTWEnv, AlfredDemangler, AlfredInfos +from ragen.env.base import BaseLanguageBasedEnv +from ragen.env.alfworld.config import AlfredEnvConfig +from ragen.env.alfworld.utils import load_config + +_RAW_ENV_CACHE = {} + +class AlfredTXTEnv(BaseLanguageBasedEnv): + + # Mode mapping: RAGEN mode -> AlfWorld train_eval + MODE_MAP = { + 'train': 'train', + 'val': None, # Will use config.eval_dataset + 'test': None, # Will use config.eval_dataset + } + + def __init__(self, config: AlfredEnvConfig = AlfredEnvConfig(), mode='train'): + """ + Initialize AlfWorld environment. + + Args: + config: AlfredEnvConfig instance + mode: 'train' or 'val' - determines which dataset split to use + """ + super().__init__() + self.config = config + self.raw_env_config = load_config(self.config.config_file) + + # Map RAGEN mode to AlfWorld train_eval + self.current_mode = mode + self._alfworld_mode = self._get_alfworld_mode(mode) + + # Initialize raw environment + self.raw_env = self._get_cached_raw_env(self._alfworld_mode) + self.num_games = self.raw_env.num_games + self.game_files = list(self.raw_env.game_files) + + self.alfred_env = None + self.current_game_file = None + self.render_cache = None + self.available_actions = None + self.render_mode = self.config.render_mode + assert self.render_mode == 'text' + + def _get_alfworld_mode(self, mode: str) -> str: + """Convert RAGEN mode to AlfWorld train_eval string.""" + if mode == 'train': + return 'train' + else: + # Use eval_dataset from config for val/test modes + return self.config.eval_dataset + + def _get_cached_raw_env(self, alfworld_mode: str) -> AlfredTWEnv: + cache_key = (os.path.abspath(self.config.config_file), alfworld_mode) + raw_env = _RAW_ENV_CACHE.get(cache_key) + if raw_env is None: + raw_env = AlfredTWEnv(config=self.raw_env_config, train_eval=alfworld_mode) + _RAW_ENV_CACHE[cache_key] = raw_env + return raw_env + + def _reinitialize_for_mode(self, mode: str): + """Reinitialize the environment when mode changes.""" + new_alfworld_mode = self._get_alfworld_mode(mode) + + if new_alfworld_mode != self._alfworld_mode: + # Close existing environment + if hasattr(self, 'alfred_env') and self.alfred_env is not None: + try: + self.alfred_env.close() + except Exception: + pass + + # Reinitialize with new mode + self._alfworld_mode = new_alfworld_mode + self.current_mode = mode + self.raw_env = self._get_cached_raw_env(self._alfworld_mode) + self.num_games = self.raw_env.num_games + self.game_files = list(self.raw_env.game_files) + self.alfred_env = None + + def reset(self, seed=None, mode=None): + """ + Reset the environment with a specific seed. + + Args: + seed: Random seed for game selection + mode: 'train', 'val', or 'test' - if different from current mode, + reinitializes the environment with the new dataset split + + Returns: + observation: Initial observation text with admissible actions + """ + try: + # Handle mode switching dynamically + if mode is not None and mode != self.current_mode: + self._reinitialize_for_mode(mode) + elif mode is not None: + self.current_mode = mode + + # Select game based on mode + if self.current_mode == "test": + if seed is None: + raise ValueError("Seed must be provided in test mode.") + selected_game = self.game_files[seed % len(self.game_files)] + else: + if seed is not None: + np.random.seed(seed) + random.seed(seed) + game_idx = seed % len(self.game_files) + selected_game = self.game_files[game_idx] + else: + selected_game = random.choice(self.game_files) + + self.current_game_file = selected_game + + if hasattr(self, 'alfred_env') and self.alfred_env is not None: + try: + self.alfred_env.close() + except Exception: + pass + + request_infos = textworld.EnvInfos(won=True, admissible_commands=True, extras=["gamefile"]) + wrappers = [AlfredDemangler(), AlfredInfos()] + max_steps = self.raw_env_config["rl"]["training"]["max_nb_steps_per_episode"] + + env_id = textworld.gym.register_game( + selected_game, + request_infos=request_infos, + batch_size=1, + asynchronous=False, + max_episode_steps=max_steps, + wrappers=wrappers + ) + + self.alfred_env = textworld.gym.make(env_id) + + obs, info = self.alfred_env.reset() + self.available_actions = info["admissible_commands"][0] + self.instruction_text = obs[0] + # Include admissible actions in render cache + self.render_cache = self._format_observation(obs[0], self.available_actions) + return self.render_cache + + except (RuntimeError, RuntimeWarning) as e: + print(f"Error in reset: {e}") + next_seed = abs(hash(str(seed))) % (2 ** 32) if seed is not None else None + return self.reset(next_seed, mode=self.current_mode) + + def _format_observation(self, obs: str, admissible_actions: list) -> str: + """Format observation to include admissible actions.""" + actions_str = ", ".join(admissible_actions) + return f"{obs}\n\nAdmissible actions: [{actions_str}]" + + def step(self, action: str): + """ + Take a step in the environment using the provided action string. + + Reward scheme (matching verl-agent): + - reward = config.score * won (10.0 on success, 0.0 otherwise) + + Args: + action: The action string to execute + + Returns: + observation: Updated observation with admissible actions + reward: Sparse reward (score on success, 0 otherwise) + done: Whether episode is finished + info: Additional information dict + """ + action_is_available = action in self.available_actions + + # Execute action in environment + obs, _, dones, infos = self.alfred_env.step([action]) + observation = obs[0] + self.available_actions = infos["admissible_commands"][0] + done = dones[0] + won = infos["won"][0] + + # Simplified reward matching verl-agent: score * won + reward = self.config.score * float(won) + + # Format observation with admissible actions + self.render_cache = self._format_observation(observation, self.available_actions) + + info = { + "action_is_effective": True, + "action_is_valid": action_is_available, + "success": won + } + + return self.render_cache, reward, done, info + + def render(self): + """Return current observation with admissible actions.""" + return self.render_cache + + def close(self): + """Clean up environment resources.""" + self.render_cache = None + if hasattr(self, 'alfred_env') and self.alfred_env is not None: + try: + self.alfred_env.close() + except Exception: + pass + + +if __name__ == "__main__": + import os + os.environ["ALFWORLD_DATA"] = os.path.expanduser("~/.cache/alfworld") + + # Test basic environment functionality + print("\n=== Test 1: Basic environment with train mode ===") + env = AlfredTXTEnv() + obs = env.reset(seed=42, mode='train') + print(f"Train observation (first 500 chars):\n{obs[:500]}...") + + # Test mode switching + print("\n=== Test 2: Mode switching to val ===") + obs = env.reset(seed=42, mode='val') + print(f"Val observation (first 500 chars):\n{obs[:500]}...") + + # Test step with admissible action + print("\n=== Test 3: Taking a step ===") + if env.available_actions: + action = env.available_actions[0] + print(f"Taking action: {action}") + obs, reward, done, info = env.step(action) + print(f"Reward: {reward}, Done: {done}, Info: {info}") + print(f"New observation (first 300 chars):\n{obs[:300]}...") + + env.close() + print("\n=== All tests completed ===") + diff --git a/ragen/eval.py b/ragen/eval.py new file mode 100644 index 0000000000000000000000000000000000000000..c95ba60ef061629acce8c5ed8839f6e833b317c5 --- /dev/null +++ b/ragen/eval.py @@ -0,0 +1,47 @@ + +from ragen.llm_agent.ctx_manager import ContextManager +from ragen.llm_agent.es_manager import EnvStateManager +from vllm import LLM, SamplingParams +from verl.single_controller.ray.base import RayWorkerGroup +from transformers import AutoTokenizer, AutoModelForCausalLM +from verl import DataProto +import hydra +import os +from typing import List, Dict +from verl.protocol import pad_dataproto_to_divisor, unpad_dataproto +from ragen.llm_agent.base_llm import ConcurrentLLM +from ragen.llm_agent.agent_proxy import ApiCallingWrapperWg, VllmWrapperWg, LLMAgentProxy, _get_rollout_do_sample + +@hydra.main(version_base=None, config_path="../config", config_name="base") +def main(config): + # detect config name from python -m ragen.llm_agent.agent_proxy --config_name frozen_lake + os.environ["VLLM_WORKER_MULTIPROC_METHOD"] = "spawn" + os.environ["CUDA_VISIBLE_DEVICES"] = str(config.system.CUDA_VISIBLE_DEVICES) + tokenizer = AutoTokenizer.from_pretrained(config.actor_rollout_ref.model.path) + actor_wg = VllmWrapperWg(config, tokenizer) + proxy = LLMAgentProxy(config, actor_wg, tokenizer) + import time + start_time = time.time() + rollouts = proxy.rollout(DataProto(batch=None, non_tensor_batch=None, meta_info={'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': _get_rollout_do_sample(config), 'validate': True}), val=True) + end_time = time.time() + print(f'rollout time: {end_time - start_time} seconds') + # print rollout rewards from the rm_scores + rm_scores = rollouts.batch["rm_scores"] + metrics = rollouts.meta_info["metrics"] + avg_reward = rm_scores.sum(-1).mean().item() + print(f'rollout rewards: {avg_reward}') + print(f'metrics:') + for k, v in metrics.items(): + print(f'{k}: {v}') + + # save results + import time as _time + timestamp = _time.strftime("%Y%m%d_%H%M%S") + save_dir = os.path.join("results", "eval") + os.makedirs(save_dir, exist_ok=True) + save_path = os.path.join(save_dir, f"val_rollouts_{timestamp}.pkl") + rollouts.save_to_disk(save_path) + print(f'save validation results to {save_path}') + +if __name__ == "__main__": + main() diff --git a/ragen/eval_api.py b/ragen/eval_api.py new file mode 100644 index 0000000000000000000000000000000000000000..d0e4b4217f081451b5495b4e800a6c885d90b0e4 --- /dev/null +++ b/ragen/eval_api.py @@ -0,0 +1,39 @@ + +from ragen.llm_agent.ctx_manager import ContextManager +from ragen.llm_agent.es_manager import EnvStateManager +from vllm import LLM, SamplingParams +from verl.single_controller.ray.base import RayWorkerGroup +from transformers import AutoTokenizer, AutoModelForCausalLM +from verl import DataProto +import hydra +import os +from typing import List, Dict +from verl.protocol import pad_dataproto_to_divisor, unpad_dataproto +from ragen.llm_agent.base_llm import ConcurrentLLM +from ragen.llm_agent.agent_proxy import ApiCallingWrapperWg, VllmWrapperWg, LLMAgentProxy + +@hydra.main(version_base=None, config_path="../config", config_name="evaluate_api_llm") +def main(config): + # detect config name from python -m ragen.llm_agent.agent_proxy --config_name frozen_lake + tokenizer = AutoTokenizer.from_pretrained(config.actor_rollout_ref.model.path) + actor_wg = ApiCallingWrapperWg(config, tokenizer) + proxy = LLMAgentProxy(config, actor_wg, tokenizer) + import time + start_time = time.time() + rollouts = proxy.rollout(DataProto(batch=None, non_tensor_batch=None, meta_info={'eos_token_id': 151645, 'pad_token_id': 151643, 'recompute_log_prob': False, 'do_sample': False, 'validate': True}), val=True) + print(f'[DEBUG] rollouts: {rollouts}') + end_time = time.time() + print(f'rollout time: {end_time - start_time} seconds') + # print rollout rewards from the rm_scores + rm_scores = rollouts.batch["rm_scores"] + metrics = rollouts.meta_info["metrics"] + avg_reward = rm_scores.sum(-1).mean().item() + print(f'rollout rewards: {avg_reward}') + print(f'metrics:') + for k, v in metrics.items(): + print(f'{k}: {v}') + + + +if __name__ == "__main__": + main() diff --git a/ragen/utils.py b/ragen/utils.py new file mode 100644 index 0000000000000000000000000000000000000000..50b6d0e26440aee66ef0812b639d900a60f26a2b --- /dev/null +++ b/ragen/utils.py @@ -0,0 +1,86 @@ +import random +import numpy as np +from contextlib import contextmanager +from omegaconf import OmegaConf +import dataclasses + +@contextmanager +def all_seed(seed): + random_state = random.getstate() + np_random_state = np.random.get_state() + + try: + random.seed(seed) + np.random.seed(seed) + yield + finally: + random.setstate(random_state) + np.random.set_state(np_random_state) + +def register_resolvers(): + try: + OmegaConf.register_new_resolver("mul", lambda x, y: x * y) + OmegaConf.register_new_resolver("int_div", lambda x, y: int(float(x) / float(y))) + OmegaConf.register_new_resolver("not", lambda x: not x) + except: + pass # already registered + + + +@dataclasses.dataclass +class GenerationsLogger: + + def log(self, loggers, samples, step, _type='val'): + if 'wandb' in loggers: + self.log_generations_to_wandb(samples, step, _type) + if 'swanlab' in loggers: + self.log_generations_to_swanlab(samples, step, _type) + + def log_generations_to_wandb(self, samples, step, _type='val'): + """Log samples to wandb as a table""" + import wandb + + # Create column names for all samples + columns = ["step"] + sum([[f"input_{i+1}", f"output_{i+1}", f"score_{i+1}"] for i in range(len(samples))], []) + + if not hasattr(self, 'table'): + # Initialize the table on first call + self.table = wandb.Table(columns=columns) + + # Create a new table with same columns and existing data + # Workaround for https://github.com/wandb/wandb/issues/2981#issuecomment-1997445737 + new_table = wandb.Table(columns=columns, data=self.table.data) + + # Add new row with all data + row_data = [] + row_data.append(step) + for sample in samples: + row_data.extend(sample) + + new_table.add_data(*row_data) + + # Update reference and log + wandb.log({f"{_type}/generations": new_table}, step=step) + self.table = new_table + + def log_generations_to_swanlab(self, samples, step, _type='val'): + """Log samples to swanlab as text""" + import swanlab + + swanlab_text_list = [] + for i, sample in enumerate(samples): + row_text = f""" + input: {sample[0]} + + --- + + output: {sample[1]} + + --- + + score: {sample[2]} + """ + swanlab_text_list.append(swanlab.Text(row_text, caption=f"sample {i+1}")) + + # Log to swanlab + swanlab.log({f"{_type}/generations": swanlab_text_list}, step=step)