config/base.yaml

defaults:
  - ppo_trainer # this is a symbolic link to the verl/verl/trainer/config/ppo_trainer.yaml file
  - envs

system:
  CUDA_VISIBLE_DEVICES: "0"

seed:
  train: 10000
  val: 123

micro_batch_size_per_gpu: 1
ppo_mini_batch_size: 16 #****
model_path: /mnt/general/share/model/Qwen/Qwen2.5-0.5B-Instruct
# /mnt/general/share/model/Qwen/Qwen2.5-0.5B-Instruct
enable_response_mask: True # Enabling response mask could improve stability of rollout/old_log_prob, as P(st|history) are no longer calculated in loss here. See https://docs.google.com/document/d/1bg7obeiKTExuHHBl5uOiSpec5uLDZ2Tgvxy6li5pHX4/edit?usp=sharing for more details.
grpo_advantage_length_weight: False # if you do not enable this and critic/advantage_estimator is GRPO, and the critic/advantages/mean is too low, then you can try enabling this to encourage reasoning and forbid collapse

lora:
  rank: 0
  alpha: 16
  target_modules: all-linear

actor_rollout_ref:
  model:
    path: ${model_path}
    lora_rank: ${lora.rank}
    lora_alpha: ${lora.alpha}
    target_modules: ${lora.target_modules}
    torch_dtype: bfloat16
  actor:
    ppo_mini_batch_size: ${ppo_mini_batch_size}  # by default, ppo_mini_batch_size = train_batch_size / 4
    ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} # following micro_batch_size_per_gpu
    use_ref: True
    entropy_coeff: 0.001
    use_kl_loss: False
    kl_loss_coef: 0.000
    kl_loss_type: kl
    clip_ratio_low: 0.2
    clip_ratio_high: 0.28
    grpo_advantage_length_weight: ${grpo_advantage_length_weight}
    optim:
      betas: [0.9, 0.999]
      lr: 1e-6
  ref:
    log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} # following micro_batch_size_per_gpu
  rollout:
    name: vllm
    log_prob_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} # following micro_batch_size_per_gpu
    tensor_model_parallel_size: 1
    max_model_len: 16384 #3600 why** 14400 
    prompt_length: 1 # useless. Just put it here
    response_length: 128 # single-turn response length 400 ****
    gpu_memory_utilization: 0.6
    max_num_batched_tokens: 16384 # set only when enable_chunked_prefill is true 
    temperature: 1
    rollout_filter_ratio: 0.25
    rollout_filter_type: largest # smallest or largest
    rollout_filter_metric: reward_variance
    enforce_eager: True #  for small models, set both enforce_eager and free_cache_engine to False to make rollout faster
    free_cache_engine: True
    val_kwargs:
      do_sample: True
      temperature: 0.5

critic:
  ppo_mini_batch_size: ${ppo_mini_batch_size} # by default, ppo_mini_batch_size = train_batch_size / 4
  ppo_micro_batch_size_per_gpu: ${micro_batch_size_per_gpu} # following micro_batch_size_per_gpu
  model:
    path: ${model_path}
    lora_rank: ${lora.rank}
    lora_alpha: ${lora.alpha}
    target_modules: ${lora.target_modules}
  optim:
    betas: [0.9, 0.999]
    lr: 1e-5

data:
  max_prompt_length: null
  max_response_length: null
  train_batch_size: null

algorithm:
  gamma: 1.0
  lam: 1.0
  high_level_gamma: 0.95
  adv_estimator: gae
  bi_level_gae: False
  kl_penalty: kl  # how to estimate kl divergence
  kl_ctrl:
    type: fixed
    kl_coef: 0.000

trainer:
  project_name: 
  experiment_name: test
  local_log_dir: "results/"
  save_freq: -1
  total_training_steps: 200
  validation_steps: 1 # validation instances = validation_steps * val_env_groups * group_size
  val_before_train: True
  n_gpus_per_node: 1
  test_freq: 10
  generations_to_log_to_wandb: 
    train: 128
    val: 20
  logger: [ 'console', 'wandb' ]
  max_actor_ckpt_to_keep: 1
  max_critic_ckpt_to_keep: 1
  default_local_dir: /mnt/general/wanghy/RAGEN/saves/

agent_proxy:
  max_context_window: -1 # set a value > 0 to enable context window for long trajectory
  max_turn: 15 #25 why** 700
  action_sep: "||"
  max_actions_per_turn: 1 # how many actions can be output at most in a single turn
  use_turn_scores: False # important to GAE when applying token-level rewards to token-level advantages. If False, will take the sum of scores as the reward for the last turn.
  enable_think: True # False -> no think RL
  reward_normalization:
    grouping: "state" # state / batch / inductive
    method: "identity" # asym_clip / identity / mean_std

es_manager:
  format_penalty: -0.1
  train:
    env_groups: 8 
    # under the same group, the env config and env seed are ensured to be equal
    group_size: 16
    env_configs:
      tags: ["CoordSokoban"]
      n_groups: [8] # If not set, all env names divide nums equally. Under the same group, the env config and env seed (prompt) are equal in each generation
  val:
    env_groups: 256
    group_size: 1 # should be set to 1 because when val temperature is set to 0 and group size > 1, there will be repetitive prompts which leads to same trajectory.
    env_configs:
      tags: ["CoordSokoban"]
      n_groups: [256] # TODO: If not set, all env names divide nums equally. Under the same group, the env config and env seed (prompt) are equal in each generation

ctx_manager:
  generation: # go to vllm
    gen_config:
      response_length: ${actor_rollout_ref.rollout.response_length}
      temperature: ${actor_rollout_ref.rollout.temperature}
      top_p: ${actor_rollout_ref.rollout.top_p}
      top_k: ${actor_rollout_ref.rollout.top_k}
      kwargs: null