Save model at global step 5

config.yaml

@@ -0,0 +1,224 @@
+data:
+  tokenizer: null
+  train_files: dapo_ds_train_sample.parquet
+  val_files: matheval.parquet
+  prompt_key: prompt
+  reward_fn_key: data_source
+  max_prompt_length: 768
+  max_response_length: 13312
+  train_batch_size: 1024
+  val_batch_size: null
+  return_raw_input_ids: false
+  return_raw_chat: false
+  shuffle: true
+  filter_overlong_prompts: true
+  filter_overlong_prompts_workers: 1
+  truncation: error
+  image_key: images
+  custom_cls:
+    path: null
+    name: null
+actor_rollout_ref:
+  hybrid_engine: true
+  model:
+    path: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
+    external_lib: null
+    override_config: {}
+    enable_gradient_checkpointing: true
+    use_remove_padding: true
+    use_liger: false
+    save_hf_repo_id: RyanYr/ppo-dapo-r1qwen1.5B-base-lr-mbs256-gpu8_actor
+    tokenizer_chat_template: null
+  actor:
+    strategy: fsdp
+    ppo_mini_batch_size: 256
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 1
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    grad_clip: 1.0
+    clip_ratio: 0.2
+    clip_ratio_low: 0.2
+    clip_ratio_high: 0.2
+    clip_ratio_c: 3.0
+    loss_agg_mode: token-mean
+    entropy_coeff: 0.001
+    use_kl_loss: false
+    use_torch_compile: true
+    kl_loss_coef: 0.001
+    kl_loss_type: low_var_kl
+    ppo_epochs: 1
+    shuffle: false
+    ulysses_sequence_parallel_size: 1
+    checkpoint:
+      contents:
+      - model
+      - optimizer
+      - extra
+    optim:
+      lr: 1.0e-06
+      lr_warmup_steps: -1
+      lr_warmup_steps_ratio: 0.0
+      min_lr_ratio: null
+      warmup_style: constant
+      total_training_steps: 100
+      weight_decay: 0.01
+    fsdp_config:
+      wrap_policy:
+        min_num_params: 0
+      param_offload: false
+      optimizer_offload: false
+      fsdp_size: -1
+  ref:
+    strategy: fsdp
+    fsdp_config:
+      param_offload: false
+      wrap_policy:
+        min_num_params: 0
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 1
+    log_prob_use_dynamic_bsz: false
+    log_prob_max_token_len_per_gpu: 16384
+    ulysses_sequence_parallel_size: 1
+  rollout:
+    name: vllm
+    temperature: 1.0
+    top_k: -1
+    top_p: 1
+    use_fire_sampling: false
+    prompt_length: 768
+    response_length: 13312
+    dtype: bfloat16
+    gpu_memory_utilization: 0.7
+    ignore_eos: false
+    enforce_eager: false
+    free_cache_engine: false
+    load_format: dummy_dtensor
+    tensor_model_parallel_size: 4
+    max_num_batched_tokens: 14080
+    max_model_len: null
+    max_num_seqs: 1024
+    log_prob_micro_batch_size: null
+    log_prob_micro_batch_size_per_gpu: 1
+    log_prob_use_dynamic_bsz: false
+    log_prob_max_token_len_per_gpu: 16384
+    disable_log_stats: true
+    enable_chunked_prefill: true
+    do_sample: true
+    'n': 1
+    engine_kwargs:
+      swap_space: null
+    val_kwargs:
+      top_k: -1
+      top_p: 1.0
+      temperature: 0
+      'n': 1
+      do_sample: false
+critic:
+  rollout_n: 1
+  strategy: fsdp
+  optim:
+    lr: 1.0e-05
+    lr_warmup_steps_ratio: 0.0
+    min_lr_ratio: null
+    warmup_style: constant
+    total_training_steps: 100
+    weight_decay: 0.01
+  model:
+    path: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
+    tokenizer_path: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
+    override_config: {}
+    external_lib: null
+    enable_gradient_checkpointing: false
+    use_remove_padding: true
+    fsdp_config:
+      param_offload: false
+      optimizer_offload: false
+      wrap_policy:
+        min_num_params: 0
+      fsdp_size: -1
+    save_hf_repo_id: RyanYr/ppo-dapo-r1qwen1.5B-base-lr-mbs256-gpu8_critic
+  ppo_mini_batch_size: 256
+  ppo_micro_batch_size: null
+  ppo_micro_batch_size_per_gpu: 1
+  forward_micro_batch_size: null
+  forward_micro_batch_size_per_gpu: 1
+  use_dynamic_bsz: false
+  ppo_max_token_len_per_gpu: 32768
+  forward_max_token_len_per_gpu: 32768
+  ulysses_sequence_parallel_size: 1
+  ppo_epochs: 1
+  shuffle: false
+  grad_clip: 1.0
+  cliprange_value: 0.5
+  checkpoint:
+    contents:
+    - model
+    - optimizer
+    - extra
+reward_model:
+  enable: false
+  strategy: fsdp
+  model:
+    input_tokenizer: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
+    path: ~/models/FsfairX-LLaMA3-RM-v0.1
+    external_lib: null
+    use_remove_padding: false
+    fsdp_config:
+      wrap_policy:
+        min_num_params: 0
+      param_offload: false
+      fsdp_size: -1
+  micro_batch_size: null
+  micro_batch_size_per_gpu: null
+  max_length: null
+  ulysses_sequence_parallel_size: 1
+  use_dynamic_bsz: false
+  forward_max_token_len_per_gpu: 32768
+  reward_manager: prime
+custom_reward_function:
+  path: null
+  name: compute_score
+algorithm:
+  gamma: 1.0
+  lam: 1.0
+  adv_estimator: gae
+  use_kl_in_reward: true
+  kl_penalty: kl
+  kl_ctrl:
+    type: fixed
+    kl_coef: 0.001
+    horizon: 10000
+    target_kl: 0.1
+trainer:
+  balance_batch: true
+  total_epochs: 100
+  total_training_steps: 100
+  project_name: value-LLM
+  experiment_name: ppo-dapo-r1qwen1.5B-base_lr-mbs256-gpu8
+  logger:
+  - console
+  - wandb
+  log_val_generations: 0
+  nnodes: 1
+  n_gpus_per_node: 8
+  save_freq: 5
+  resume_mode: auto
+  resume_from_path: null
+  val_before_train: false
+  test_freq: -1
+  critic_warmup: 0
+  default_hdfs_dir: null
+  del_local_ckpt_after_load: false
+  default_local_dir: checkpoints/value-LLM/ppo-dapo-r1qwen1.5B-base_lr-mbs256-gpu8
+  max_actor_ckpt_to_keep: 1
+  max_critic_ckpt_to_keep: 1
+  ray_wait_register_center_timeout: 300
+  hf_token: null
+  resume_from_hf:
+    enable: false
+    actor_hf_repo_id: null
+    actor_revision: main
+    critic_hf_repo_id: null
+    critic_revision: main
+    hf_token: null

extra_state_world_size_8_rank_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:951e3e58f6d18c9067d36cb378af10e988c7c69bf2d01101104207385526abd3
+size 14632

extra_state_world_size_8_rank_1.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:553368d290906e456c7e7d9e864926a888564dfb09dedcef332fb24fb7d689cd
+size 14632

extra_state_world_size_8_rank_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:604929a7af71e743598b3597a38aec6ba0a0f3823b3765dd917014624220ec9c
+size 14632

extra_state_world_size_8_rank_3.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:895572e6f7a7db8dd4d25277ebd2a9bcb3f7df1e1cd9db605607f17e421151b0
+size 14632

extra_state_world_size_8_rank_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6f361a38800d8aa7dcc2b70ca9ac0ea2a7e50965aaf257697b551e1b860faaf6
+size 14632

extra_state_world_size_8_rank_5.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:335d9ec86bd46255f3393e37dcdee25f15edc6a348123f5ef34c527441fedb8e
+size 14632

extra_state_world_size_8_rank_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47d6fa2bff9167262da817374d5e4237b2490287ace82fd3ca0ccb424b997c2a
+size 14632

extra_state_world_size_8_rank_7.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:87854d2eafbc04a1ac2cba98a983e77b21ba246b8adfb54f5da28e38c5299198
+size 14632

model_world_size_8_rank_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e17212872b49953432f6cd5c20fa0f85a0b27dad01e2c132ef7aa129afd61541
+size 772052570

model_world_size_8_rank_1.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a1149f1e9624c2c6ac5120f5cd95cca3e7c99dae0df7ac45f691334c18dc612e
+size 772046426

model_world_size_8_rank_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:942a3aa8b91b6e30d1284d7d904c3567051a249111d25d598bf4f0917646e9e8
+size 772046426

model_world_size_8_rank_3.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1f5afb32b9b46c70a3d2b4b47442475b1bb04112a0bfc11202e3fec80650ab4
+size 772046426

model_world_size_8_rank_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cbdb4c838d77bca7f43e621502a2c7118f5c6a0950fa20f7530cd794f487d455
+size 772046426

model_world_size_8_rank_5.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:01375b9e5bf1f25dd3f55823ee6c15cec9a671c378a30c265500a1d70f77e169
+size 772046426

model_world_size_8_rank_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:10e74a60b5bf022c3acf956ba9ae04957640213d9079720817fad5ba56c0907e
+size 772046426

model_world_size_8_rank_7.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8afd83c39978e545e58572b69507bf7aa9fffb90a856df3c1e5fa34ec3d27afc
+size 772046426

optim_world_size_8_rank_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d958c9490ff952a6e9e71cfe33d928bcc871718c21b5cad6858378219569169a
+size 1543742007

optim_world_size_8_rank_1.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cdb1f1af365e88aefd75343cc45876b5a218575a1108fb1ad474fe99fa2b75cf
+size 1543742007

optim_world_size_8_rank_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd39afb3a442d339bee2c67c778d8be24934ecd763d09af8f035af9d852cbdbf
+size 1543742007

optim_world_size_8_rank_3.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a521c6b08e69c1bbe2fa781c3974660239a5e675b07751cae21457891782bd13
+size 1543742007

optim_world_size_8_rank_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d60653894255cc6cfac8b4afe016a7edb51a71afd2327fc414e60c9ac7fd8756
+size 1543742007

optim_world_size_8_rank_5.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac53bc6b6921e1040d72a8bb545435388943737309e29340165d6447bab15bf4
+size 1543742007

optim_world_size_8_rank_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:270959f667768fca5ff40876f60dbb92f380309462cd61337e9a6d9955debbb7
+size 1543742007

optim_world_size_8_rank_7.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:013c7c72cd28d277913b4b27f9b1692c440926b931734c8bacf9beac0fe8392b
+size 1543742007