Save model at global step 5

config.yaml

@@ -0,0 +1,178 @@
+data:
+  tokenizer: null
+  train_files: ace-numina_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: 640
+  return_raw_input_ids: false
+  return_raw_chat: false
+  shuffle: true
+  filter_overlong_prompts: false
+  filter_overlong_prompts_workers: 1
+  filter_overlong_responses: true
+  truncation: left
+  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: false
+    use_liger: false
+    save_hf_repo_id: RyanYr/brm-ace-numina-r1qwen1.5B-base-lr2.5e-6-beta0.002
+    tokenizer_chat_template: null
+  actor:
+    brm:
+      norm_factor: value
+      value_constant: null
+    buffer:
+      buffer_type: null
+      offline_dataset_buffer:
+        train_files: null
+        response_key: response
+        response_truncation: right
+        shuffle: true
+        update_size: 1024
+    strategy: fsdp
+    ppo_mini_batch_size: 1024
+    ppo_micro_batch_size: null
+    ppo_micro_batch_size_per_gpu: 2
+    use_dynamic_bsz: false
+    ppo_max_token_len_per_gpu: 16384
+    grad_clip: 1.0
+    use_torch_compile: true
+    ppo_epochs: 1
+    shuffle: false
+    ulysses_sequence_parallel_size: 1
+    checkpoint:
+      contents:
+      - model
+      - optimizer
+      - extra
+    optim:
+      lr: 2.5e-06
+      lr_warmup_steps: -1
+      lr_warmup_steps_ratio: 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
+    report_entropy: false
+  ref:
+    ref_model_path: deepseek-ai/DeepSeek-R1-Distill-Qwen-1.5B
+    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: 4
+    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: 0.8
+    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: 2
+    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: 0.6
+      'n': 32
+      do_sample: true
+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: 16384
+  reward_manager: prime
+  reward_kwargs:
+    format_reward: 0.0
+    format_type: null
+custom_reward_function:
+  path: null
+  name: compute_score
+trainer:
+  balance_batch: true
+  total_epochs: 100
+  total_training_steps: 100
+  project_name: value-LLM
+  experiment_name: brm-ace-numina-r1qwen1.5B-base_lr2.5e-6-beta0.002
+  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
+  default_hdfs_dir: null
+  del_local_ckpt_after_load: false
+  default_local_dir: ./BRM
+  max_actor_ckpt_to_keep: 1
+  ray_wait_register_center_timeout: 300
+  hf_token: null
+  resume_from_hf:
+    enable: false
+    hf_repo_id: null
+    hf_token: null
+    revision: main
+algorithm:
+  kl_ctrl:
+    kl_coef: 0.002

data.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a0f9666826cfac8d4b325eaf87633050dedf8c3d29ea2e341f9392ded4e7b81e
+size 1492

extra_state_world_size_8_rank_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2ba5748c2912062a3d03558edf4e18d19f7cdbe792c8157e87cbf0c4733bdcdc
+size 14632

extra_state_world_size_8_rank_1.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c45b43345c7bfb5f859ec8c657bd63c4488c85e1153b6a755e0fa49873011e7
+size 14632

extra_state_world_size_8_rank_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:38b8a507cbc0aad8f79aabaad028adba3e91ed06590e2521c679a73609b3cbfe
+size 14632

extra_state_world_size_8_rank_3.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:60acb24f0a5933736d62ac943458714910a4e2f8114ebf9464b03146b1a04621
+size 14632

extra_state_world_size_8_rank_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0b2925a2b769fce6505d3902718c9351851287f8dd8b719318da2654a0571a4b
+size 14632

extra_state_world_size_8_rank_5.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8f12794f2a8e8e04e493bd334a98bd451dcc5b355623af4c47209880046765fc
+size 14632

extra_state_world_size_8_rank_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e11319d98ccb1476b0cdedab864a6d8fdd6af32dede866c8d0386f35d8edd55
+size 14632

extra_state_world_size_8_rank_7.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f34da93ba77dd57f3133991fc7b78c8605fa74124807f34171d12710e2eea99f
+size 14632

model_world_size_8_rank_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c2c0c654d0cbc9f4ddfe178eefee8ffebc04162b5ec96c27a36fa3bf213f2fe7
+size 888732218

model_world_size_8_rank_1.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:81bee9d4a9f23cbbadb33f8918aef85a582e0410980f3051088cd45d2b326d8e
+size 888732218

model_world_size_8_rank_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f24221a0acf892b7f6393aae0326d06baca4022e9e53f362905a101b49c0a7ca
+size 888732218

model_world_size_8_rank_3.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:47546e97f815cf490450ffe79d3a7024f3321e00d608a431321be231a3ec062c
+size 888732218

model_world_size_8_rank_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7ead6a94d40eee14e3cb53dfa14da440f3a9e9b4c9323d6a01ec2b7a1878b36
+size 888732218

model_world_size_8_rank_5.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dd50bb9ff0ff2c108e3c32783c31b31c43a8ce6a13cc948f1dbf4b5b078ac5a1
+size 888732218

model_world_size_8_rank_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bbd9b6770aa83e294e5f62dd853b7c9e8201279f67c5864093442256190a50fb
+size 888732218

model_world_size_8_rank_7.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9fe5b9b1d9fa7a7223c37fbb34fb113faadde5c2ea5550f14fec73cb1e6fb081
+size 888732218

optim_world_size_8_rank_0.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:21a8e411dc546ae4692659d826f7bcab9862a34a8dc4f77247854d4f4f2c5877
+size 1777114167

optim_world_size_8_rank_1.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:63d4fb85e9052994148c681e404ac93db2c4299c5f58df81d12ea3fad67b90bf
+size 1777114167

optim_world_size_8_rank_2.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9f377c493bc413461ba645b1ae65e4e748e7c0865e8afc7e424663e7903f3e06
+size 1777114167

optim_world_size_8_rank_3.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:845d9d44b8d40fabf2bce0330cc3d18cc2d9c50938a9697fe82c072731eda1ac
+size 1777114167

optim_world_size_8_rank_4.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:537a4144e1d257e41a15c9b919428f642fd2ed5ca479b1a0930d20fc86db2f7f
+size 1777114167

optim_world_size_8_rank_5.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:64cf7a048eb4a0110df1ee001ef0084b335104064e62d9ae4a9e547d83558392
+size 1777114167

optim_world_size_8_rank_6.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a293e83ce6eca09899776bb843795cf312a619d78583c38cd6b406a77597d0b5
+size 1777114167

optim_world_size_8_rank_7.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3ec4e3cc0cc7ea4487d597906242e9c8bc49fc40e34aeb83d7fed866e30942bc
+size 1777114167