Add files using upload-large-folder tool

code/RL_model/verl/verl_train/outputs/2026-02-01/08-00-20/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 256
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 32
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 32
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    fsdp_config:
+      _target_: verl.workers.config.FSDPEngineConfig
+      wrap_policy:
+        min_num_params: 0
+      param_offload: true
+      optimizer_offload: false
+      offload_policy: false
+      reshard_after_forward: true
+      fsdp_size: -1
+      forward_prefetch: false
+      model_dtype: fp32
+      use_orig_params: false
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.6
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 2
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 32
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: verl
+  experiment_name: qwen3-4b-instruct-2507-function-rm
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/08-00-20/.hydra/hydra.yaml

@@ -0,0 +1,204 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=256
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=32
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=2
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.6
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32
+    - actor_rollout_ref.ref.fsdp_config.param_offload=True
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=verl
+    - trainer.experiment_name=qwen3-4b-instruct-2507-function-rm
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=32,actor_rollout_ref.actor.ppo_mini_batch_size=256,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=True,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32,actor_rollout_ref.rollout.gpu_memory_utilization=0.6,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=2,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-2507-function-rm,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl/outputs/2026-02-01/08-00-20
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/08-00-20/.hydra/overrides.yaml

@@ -0,0 +1,37 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=256
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=32
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32
+- actor_rollout_ref.rollout.tensor_model_parallel_size=2
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.6
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32
+- actor_rollout_ref.ref.fsdp_config.param_offload=True
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=verl
+- trainer.experiment_name=qwen3-4b-instruct-2507-function-rm
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/08-03-02/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 256
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 32
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 32
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    fsdp_config:
+      _target_: verl.workers.config.FSDPEngineConfig
+      wrap_policy:
+        min_num_params: 0
+      param_offload: true
+      optimizer_offload: false
+      offload_policy: false
+      reshard_after_forward: true
+      fsdp_size: -1
+      forward_prefetch: false
+      model_dtype: fp32
+      use_orig_params: false
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.6
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 2
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 32
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: verl
+  experiment_name: qwen3-4b-instruct-2507-function-rm
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/08-03-02/.hydra/hydra.yaml

@@ -0,0 +1,204 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=256
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=32
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=2
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.6
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32
+    - actor_rollout_ref.ref.fsdp_config.param_offload=True
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=verl
+    - trainer.experiment_name=qwen3-4b-instruct-2507-function-rm
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=32,actor_rollout_ref.actor.ppo_mini_batch_size=256,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=True,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32,actor_rollout_ref.rollout.gpu_memory_utilization=0.6,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=2,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-2507-function-rm,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl/outputs/2026-02-01/08-03-02
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/08-03-02/.hydra/overrides.yaml

@@ -0,0 +1,37 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=256
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=32
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=32
+- actor_rollout_ref.rollout.tensor_model_parallel_size=2
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.6
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=32
+- actor_rollout_ref.ref.fsdp_config.param_offload=True
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=verl
+- trainer.experiment_name=qwen3-4b-instruct-2507-function-rm
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/08-12-45/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: verl
+  experiment_name: qwen3-4b-instruct-optimized
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/08-12-45/.hydra/hydra.yaml

@@ -0,0 +1,204 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl/outputs/2026-02-01/08-12-45
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/08-12-45/.hydra/overrides.yaml

@@ -0,0 +1,37 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/11-18-25/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: verl
+  experiment_name: qwen3-4b-instruct-optimized
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/11-18-25/.hydra/hydra.yaml

@@ -0,0 +1,204 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/11-18-25
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/11-18-25/.hydra/overrides.yaml

@@ -0,0 +1,37 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/11-27-06/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: verl
+  experiment_name: qwen3-4b-instruct-optimized
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/11-27-06/.hydra/hydra.yaml

@@ -0,0 +1,204 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/11-27-06
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/11-27-06/.hydra/overrides.yaml

@@ -0,0 +1,37 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/11-29-26/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: verl
+  experiment_name: qwen3-4b-instruct-optimized
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/11-29-26/.hydra/hydra.yaml

@@ -0,0 +1,204 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/11-29-26
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/11-29-26/.hydra/overrides.yaml

@@ -0,0 +1,37 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/11-47-36/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 256
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: null
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: false
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: null
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: ???
+    mode: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 2
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: null
+    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: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: ~/models/deepseek-llm-7b-chat
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 512
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 30
+  total_training_steps: null
+  project_name: verl_examples
+  experiment_name: gsm8k
+  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: -1
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/11-47-36/.hydra/hydra.yaml

@@ -0,0 +1,170 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: algorithm.adv_estimator=grpo,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/11-47-36
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/11-47-36/.hydra/overrides.yaml

@@ -0,0 +1,3 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet

code/RL_model/verl/verl_train/outputs/2026-02-01/11-58-50/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 256
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: null
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: false
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: null
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: ???
+    mode: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 2
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: null
+    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: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: ~/models/deepseek-llm-7b-chat
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 512
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 30
+  total_training_steps: null
+  project_name: verl_examples
+  experiment_name: gsm8k
+  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: -1
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/11-58-50/.hydra/hydra.yaml

@@ -0,0 +1,170 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: algorithm.adv_estimator=grpo,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/11-58-50
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/11-58-50/.hydra/overrides.yaml

@@ -0,0 +1,3 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet

code/RL_model/verl/verl_train/outputs/2026-02-01/12-06-58/.hydra/config.yaml

@@ -0,0 +1,651 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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:
+    strategy: remote
+    custom_reward_function:
+      path: /home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: verl
+  experiment_name: qwen3-4b-instruct-optimized
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/12-06-58/.hydra/hydra.yaml

@@ -0,0 +1,206 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - +reward_model.reward_manager.strategy=remote
+    - +reward_model.reward_manager.custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: +reward_model.reward_manager.custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py,+reward_model.reward_manager.strategy=remote,actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/12-06-58
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/12-06-58/.hydra/overrides.yaml

@@ -0,0 +1,39 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- +reward_model.reward_manager.strategy=remote
+- +reward_model.reward_manager.custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/21-08-13/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: /home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: readctrl-verl
+  experiment_name: qwen3-4b-instruct-optimized-multiclinsum-gs
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/21-08-13/.hydra/hydra.yaml

@@ -0,0 +1,205 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=readctrl-verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=readctrl-verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/21-08-13
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/21-08-13/.hydra/overrides.yaml

@@ -0,0 +1,38 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=readctrl-verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/21-10-05/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: /home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: readctrl-verl
+  experiment_name: qwen3-4b-instruct-optimized-multiclinsum-gs
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/21-10-05/.hydra/hydra.yaml

@@ -0,0 +1,205 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=readctrl-verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=readctrl-verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/21-10-05
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/21-10-05/.hydra/overrides.yaml

@@ -0,0 +1,38 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=readctrl-verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/21-14-31/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: /home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: readctrl-verl
+  experiment_name: qwen3-4b-instruct-optimized-multiclinsum-gs
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/21-14-31/.hydra/hydra.yaml

@@ -0,0 +1,205 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=readctrl-verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=readctrl-verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/21-14-31
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/21-14-31/.hydra/overrides.yaml

@@ -0,0 +1,38 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=readctrl-verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/21-36-33/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: /home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: readctrl-verl
+  experiment_name: qwen3-4b-instruct-optimized-multiclinsum-gs
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/21-36-33/.hydra/hydra.yaml

@@ -0,0 +1,205 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=readctrl-verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=readctrl-verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/21-36-33
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/21-36-33/.hydra/overrides.yaml

@@ -0,0 +1,38 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=readctrl-verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/22-02-10/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: /home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: readctrl-verl
+  experiment_name: qwen3-4b-instruct-optimized-multiclinsum-gs
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/22-02-10/.hydra/hydra.yaml

@@ -0,0 +1,205 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=readctrl-verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=readctrl-verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/22-02-10
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/22-02-10/.hydra/overrides.yaml

@@ -0,0 +1,38 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=readctrl-verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15

code/RL_model/verl/verl_train/outputs/2026-02-01/22-08-56/.hydra/config.yaml

@@ -0,0 +1,648 @@
+actor_rollout_ref:
+  actor:
+    optim:
+      _target_: verl.workers.config.FSDPOptimizerConfig
+      optimizer: AdamW
+      optimizer_impl: torch.optim
+      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
+      lr_scheduler_type: constant
+      warmup_style: null
+      override_optimizer_config: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    strategy: fsdp
+    ppo_mini_batch_size: 512
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 64
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    tau_pos: 1.0
+    tau_neg: 1.05
+    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
+    loss_scale_factor: null
+    entropy_coeff: 0
+    calculate_entropy: false
+    use_kl_loss: true
+    use_prefix_grouper: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    data_loader_seed: 42
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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}
+    calculate_sum_pi_squared: false
+    sum_pi_squared_checkpointing: false
+  ref:
+    rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+    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: 64
+    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: level0
+          analysis: true
+          discrete: false
+        torch:
+          _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+          contents: []
+          discrete: false
+        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}
+    router_replay:
+      _target_: verl.workers.config.RouterReplayConfig
+      mode: disabled
+      record_file: null
+      replay_file: null
+    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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: true
+      strategy: fsdp
+      dtype: bfloat16
+    _target_: verl.workers.config.FSDPActorConfig
+    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: async
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    prompt_length: ${oc.select:data.max_prompt_length,512}
+    response_length: ${oc.select:data.max_response_length,512}
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    tensor_model_parallel_size: 1
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    pipeline_model_parallel_size: 1
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    logprobs_mode: processed_logprobs
+    scheduling_policy: fcfs
+    load_format: dummy
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 64
+    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': 5
+    over_sample_rate: 0
+    multi_stage_wake_up: false
+    engine_kwargs:
+      vllm: {}
+      sglang: {}
+      trtllm: {}
+    val_kwargs:
+      _target_: verl.workers.config.SamplingConfig
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+    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
+      default_agent_loop: single_turn_agent
+      agent_loop_config_path: null
+      custom_async_server:
+        _target_: verl.workers.config.CustomAsyncServerConfig
+        path: null
+        name: null
+    checkpoint_engine:
+      _target_: verl.workers.config.CheckpointEngineConfig
+      backend: naive
+      update_weights_bucket_megabytes: 2048
+      engine_kwargs: {}
+    trace:
+      _target_: verl.workers.config.TraceConfig
+      backend: null
+      token2text: false
+      max_samples_per_step_per_worker: null
+    skip_rollout: false
+    skip_dump_dir: /tmp/rollout_dump
+    skip_tokenizer_init: true
+    enable_rollout_routing_replay: false
+    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}
+    prometheus:
+      _target_: verl.workers.config.PrometheusConfig
+      enable: false
+      port: 9090
+      file: /tmp/ray/session_latest/metrics/prometheus/prometheus.yml
+      served_model_name: ${oc.select:actor_rollout_ref.model.path,null}
+    quantization: null
+    quantization_config_file: null
+    mtp: ${oc.select:actor_rollout_ref.model.mtp, null}
+    layered_summon: false
+  model:
+    _target_: verl.workers.config.HFModelConfig
+    path: Qwen/Qwen3-4B-Instruct-2507
+    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: true
+    lora_rank: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    exclude_modules: null
+    lora_adapter_path: null
+    use_liger: false
+    use_fused_kernels: false
+    fused_kernel_options:
+      impl_backend: torch
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+    mtp:
+      _target_: verl.workers.config.MtpConfig
+      enable: false
+      enable_train: false
+      enable_rollout: false
+      detach_encoder: false
+      mtp_loss_scaling_factor: 0.1
+      speculative_algorithm: EAGLE
+      speculative_num_steps: 3
+      speculative_eagle_topk: 1
+      speculative_num_draft_tokens: 4
+      method: mtp
+      num_speculative_tokens: 1
+  hybrid_engine: true
+  nccl_timeout: 600
+data:
+  tokenizer: null
+  use_shm: false
+  train_files: /home/mshahidul/data/gsm8k/train.parquet
+  val_files: /home/mshahidul/data/gsm8k/test.parquet
+  train_max_samples: -1
+  val_max_samples: -1
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 512
+  max_response_length: 1024
+  train_batch_size: 1024
+  val_batch_size: null
+  tool_config_path: ${oc.select:actor_rollout_ref.rollout.multi_turn.tool_config_path,
+    null}
+  return_raw_input_ids: false
+  return_raw_chat: true
+  return_full_prompt: false
+  shuffle: true
+  seed: null
+  dataloader_num_workers: 8
+  image_patch_size: 14
+  validation_shuffle: false
+  filter_overlong_prompts: true
+  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: {}
+reward_manager:
+  _target_: verl.trainer.config.config.RewardManagerConfig
+  source: register
+  name: ${oc.select:reward_model.reward_manager,naive}
+  module:
+    _target_: verl.trainer.config.config.ModuleConfig
+    path: null
+    name: custom_reward_manager
+critic:
+  optim:
+    _target_: verl.workers.config.FSDPOptimizerConfig
+    optimizer: AdamW
+    optimizer_impl: torch.optim
+    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
+    lr_scheduler_type: constant
+    warmup_style: null
+    override_optimizer_config: null
+  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
+      seed: 42
+      full_determinism: false
+      ulysses_sequence_parallel_size: 1
+      entropy_from_logits_with_chunking: false
+      use_torch_compile: true
+      entropy_checkpointing: false
+      forward_only: false
+      strategy: fsdp
+      dtype: bfloat16
+    path: ~/models/deepseek-llm-7b-chat
+    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: 0
+    lora_alpha: 16
+    target_modules: all-linear
+    tiled_mlp:
+      enabled: false
+      num_shards: 4
+  _target_: verl.workers.config.FSDPCriticConfig
+  rollout_n: ${oc.select:actor_rollout_ref.rollout.n,1}
+  strategy: fsdp
+  enable: null
+  ppo_mini_batch_size: ${oc.select:actor_rollout_ref.actor.ppo_mini_batch_size,256}
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: ${oc.select:.ppo_micro_batch_size,null}
+  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}
+  data_loader_seed: 42
+  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: level0
+        analysis: true
+        discrete: false
+      torch:
+        _target_: verl.utils.profiler.config.TorchProfilerToolConfig
+        contents: []
+        discrete: false
+      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: 8
+  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
+    override_config: {}
+    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
+  reward_loop_source: register
+  reward_loop_module_path: null
+  reward_loop_class_name: null
+  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
+  use_reward_loop: true
+  num_workers: 1
+  rollout:
+    _target_: verl.workers.config.RolloutConfig
+    name: ???
+    dtype: bfloat16
+    gpu_memory_utilization: 0.5
+    enforce_eager: true
+    cudagraph_capture_sizes: null
+    free_cache_engine: true
+    data_parallel_size: 1
+    expert_parallel_size: 1
+    tensor_model_parallel_size: 2
+    max_num_batched_tokens: 8192
+    max_model_len: null
+    max_num_seqs: 1024
+    load_format: auto
+    engine_kwargs: {}
+    limit_images: null
+    enable_chunked_prefill: true
+    enable_prefix_caching: true
+    disable_log_stats: true
+    skip_tokenizer_init: false
+    prompt_length: 2048
+    response_length: 2048
+algorithm:
+  rollout_correction:
+    rollout_is: null
+    rollout_is_threshold: 2.0
+    rollout_rs: null
+    rollout_rs_threshold: null
+    bypass_mode: false
+    loss_type: ppo_clip
+    rollout_is_batch_normalize: false
+  _target_: verl.trainer.config.AlgoConfig
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: grpo
+  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: 0.001
+    horizon: 10000
+    target_kl: 0.1
+  use_pf_ppo: false
+  pf_ppo:
+    reweight_method: pow
+    weight_pow: 2.0
+custom_reward_function:
+  path: /home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 15
+  total_training_steps: null
+  project_name: readctrl-verl
+  experiment_name: qwen3-4b-instruct-optimized-multiclinsum-gs
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  rollout_data_dir: null
+  validation_data_dir: null
+  nnodes: 1
+  n_gpus_per_node: 2
+  save_freq: 20
+  esi_redundant_time: 0
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: true
+  val_only: false
+  test_freq: 5
+  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: null
+  max_critic_ckpt_to_keep: null
+  ray_wait_register_center_timeout: 300
+  device: cuda
+  use_legacy_worker_impl: auto
+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: {}
+transfer_queue:
+  enable: false
+ray_kwargs:
+  ray_init:
+    num_cpus: null
+  timeline_json_file: null

code/RL_model/verl/verl_train/outputs/2026-02-01/22-08-56/.hydra/hydra.yaml

@@ -0,0 +1,205 @@
+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:
+    - algorithm.adv_estimator=grpo
+    - data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+    - data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+    - custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+    - data.train_batch_size=1024
+    - data.max_prompt_length=512
+    - data.max_response_length=1024
+    - data.filter_overlong_prompts=True
+    - data.truncation=error
+    - actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+    - actor_rollout_ref.actor.optim.lr=1e-6
+    - actor_rollout_ref.model.use_remove_padding=True
+    - actor_rollout_ref.actor.ppo_mini_batch_size=512
+    - actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.actor.use_kl_loss=True
+    - actor_rollout_ref.actor.kl_loss_coef=0.001
+    - actor_rollout_ref.actor.kl_loss_type=low_var_kl
+    - actor_rollout_ref.actor.entropy_coeff=0
+    - actor_rollout_ref.model.enable_gradient_checkpointing=True
+    - actor_rollout_ref.actor.fsdp_config.param_offload=False
+    - actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+    - actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.rollout.tensor_model_parallel_size=1
+    - actor_rollout_ref.rollout.name=vllm
+    - actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+    - actor_rollout_ref.rollout.n=5
+    - actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+    - actor_rollout_ref.ref.fsdp_config.param_offload=False
+    - algorithm.use_kl_in_reward=False
+    - trainer.critic_warmup=0
+    - trainer.logger=["console","wandb"]
+    - trainer.project_name=readctrl-verl
+    - trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+    - trainer.n_gpus_per_node=2
+    - trainer.nnodes=1
+    - trainer.save_freq=20
+    - trainer.test_freq=5
+    - trainer.total_epochs=15
+  job:
+    name: main_ppo
+    chdir: null
+    override_dirname: actor_rollout_ref.actor.entropy_coeff=0,actor_rollout_ref.actor.fsdp_config.optimizer_offload=False,actor_rollout_ref.actor.fsdp_config.param_offload=False,actor_rollout_ref.actor.kl_loss_coef=0.001,actor_rollout_ref.actor.kl_loss_type=low_var_kl,actor_rollout_ref.actor.optim.lr=1e-6,actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64,actor_rollout_ref.actor.ppo_mini_batch_size=512,actor_rollout_ref.actor.use_kl_loss=True,actor_rollout_ref.model.enable_gradient_checkpointing=True,actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507,actor_rollout_ref.model.use_remove_padding=True,actor_rollout_ref.ref.fsdp_config.param_offload=False,actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.gpu_memory_utilization=0.7,actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64,actor_rollout_ref.rollout.n=5,actor_rollout_ref.rollout.name=vllm,actor_rollout_ref.rollout.tensor_model_parallel_size=1,algorithm.adv_estimator=grpo,algorithm.use_kl_in_reward=False,custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py,data.filter_overlong_prompts=True,data.max_prompt_length=512,data.max_response_length=1024,data.train_batch_size=1024,data.train_files=/home/mshahidul/data/gsm8k/train.parquet,data.truncation=error,data.val_files=/home/mshahidul/data/gsm8k/test.parquet,trainer.critic_warmup=0,trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs,trainer.logger=["console","wandb"],trainer.n_gpus_per_node=2,trainer.nnodes=1,trainer.project_name=readctrl-verl,trainer.save_freq=20,trainer.test_freq=5,trainer.total_epochs=15
+    id: ???
+    num: ???
+    config_name: ppo_trainer
+    env_set: {}
+    env_copy: []
+    config:
+      override_dirname:
+        kv_sep: '='
+        item_sep: ','
+        exclude_keys: []
+  runtime:
+    version: 1.3.2
+    version_base: '1.3'
+    cwd: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train
+    config_sources:
+    - path: hydra.conf
+      schema: pkg
+      provider: hydra
+    - path: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/verl/trainer/config
+      schema: file
+      provider: main
+    - path: ''
+      schema: structured
+      provider: schema
+    output_dir: /data/home_beta/mshahidul/readctrl/code/RL_model/verl/verl_train/outputs/2026-02-01/22-08-56
+    choices:
+      algorithm@algorithm.rollout_correction: rollout_correction
+      reward_model: dp_reward_loop
+      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

code/RL_model/verl/verl_train/outputs/2026-02-01/22-08-56/.hydra/overrides.yaml

@@ -0,0 +1,38 @@
+- algorithm.adv_estimator=grpo
+- data.train_files=/home/mshahidul/data/gsm8k/train.parquet
+- data.val_files=/home/mshahidul/data/gsm8k/test.parquet
+- custom_reward_function.path=/home/mshahidul/readctrl/code/RL_model/verl/verl_train/reward_func/reward.py
+- data.train_batch_size=1024
+- data.max_prompt_length=512
+- data.max_response_length=1024
+- data.filter_overlong_prompts=True
+- data.truncation=error
+- actor_rollout_ref.model.path=Qwen/Qwen3-4B-Instruct-2507
+- actor_rollout_ref.actor.optim.lr=1e-6
+- actor_rollout_ref.model.use_remove_padding=True
+- actor_rollout_ref.actor.ppo_mini_batch_size=512
+- actor_rollout_ref.actor.ppo_micro_batch_size_per_gpu=64
+- actor_rollout_ref.actor.use_kl_loss=True
+- actor_rollout_ref.actor.kl_loss_coef=0.001
+- actor_rollout_ref.actor.kl_loss_type=low_var_kl
+- actor_rollout_ref.actor.entropy_coeff=0
+- actor_rollout_ref.model.enable_gradient_checkpointing=True
+- actor_rollout_ref.actor.fsdp_config.param_offload=False
+- actor_rollout_ref.actor.fsdp_config.optimizer_offload=False
+- actor_rollout_ref.rollout.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.rollout.tensor_model_parallel_size=1
+- actor_rollout_ref.rollout.name=vllm
+- actor_rollout_ref.rollout.gpu_memory_utilization=0.7
+- actor_rollout_ref.rollout.n=5
+- actor_rollout_ref.ref.log_prob_micro_batch_size_per_gpu=64
+- actor_rollout_ref.ref.fsdp_config.param_offload=False
+- algorithm.use_kl_in_reward=False
+- trainer.critic_warmup=0
+- trainer.logger=["console","wandb"]
+- trainer.project_name=readctrl-verl
+- trainer.experiment_name=qwen3-4b-instruct-optimized-multiclinsum-gs
+- trainer.n_gpus_per_node=2
+- trainer.nnodes=1
+- trainer.save_freq=20
+- trainer.test_freq=5
+- trainer.total_epochs=15