Add step_25

.gitattributes

@@ -33,3 +33,21 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/.metadata filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__0_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__10_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__11_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__12_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__13_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__14_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__15_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__1_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__2_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__3_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__4_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__5_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__6_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__7_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__8_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/optimizer/optim/__9_0.distcp filter=lfs diff=lfs merge=lfs -text
+step_25/policy/tokenizer/tokenizer.json filter=lfs diff=lfs merge=lfs -text

step_25/config.yaml

@@ -0,0 +1,394 @@
+checkpointing:
+  checkpoint_dir: /p/scratch/scifi/kryeziu1/opsd/checkpoints/distill-be-concise-4b-self
+  checkpoint_must_save_by: null
+  enabled: true
+  higher_is_better: true
+  keep_top_k: 3
+  metric_name: val:accuracy
+  model_save_format: safetensors
+  save_consolidated: false
+  save_period: 25
+cluster:
+  gpus_per_node: 4
+  num_nodes: 4
+data:
+  default:
+    env_name: math
+    prompt_file: examples/prompts/student-concise.txt
+    system_prompt_file: null
+    teacher_prompt_file: examples/prompts/teacher-concise.txt
+  max_input_seq_length: 17408
+  shuffle: true
+  train:
+  - dataset_name: OpenThoughtsMath
+    env_name: math
+    prompt_file: examples/prompts/student-concise.txt
+    system_prompt_file: null
+    teacher_prompt_file: examples/prompts/teacher-concise.txt
+  validation:
+  - dataset_name: math500
+    env_name: math
+    prompt_file: examples/prompts/cot.txt
+    repeat: 4
+    system_prompt_file: null
+    teacher_prompt_file: null
+  - dataset_name: AIME2024
+    env_name: math
+    prompt_file: examples/prompts/cot.txt
+    repeat: 4
+    system_prompt_file: null
+    teacher_prompt_file: null
+  - dataset_name: AIME2025
+    env_name: math
+    prompt_file: examples/prompts/cot.txt
+    repeat: 4
+    system_prompt_file: null
+    teacher_prompt_file: null
+  - dataset_name: AIME2026
+    env_name: math
+    prompt_file: examples/prompts/cot.txt
+    repeat: 4
+    system_prompt_file: null
+    teacher_prompt_file: null
+distillation:
+  max_num_epochs: 1
+  max_num_steps: 1000
+  max_rollout_turns: 1
+  max_val_samples: 2560
+  num_generations_per_prompt: 1
+  num_prompts_per_step: 64
+  seed: 42
+  teacher_student_prefix_fraction: 0.0
+  topk_logits_k: 512
+  val_at_end: false
+  val_at_start: true
+  val_batch_size: 512
+  val_max_new_tokens: 32768
+  val_max_total_sequence_length: 33280
+  val_period: 25
+  val_steps:
+  - 10
+  val_temperature: 0.6
+env:
+  math:
+    num_workers: 8
+logger:
+  gpu_monitoring:
+    collection_interval: 10
+    flush_interval: 10
+  log_dir: logs/distill-be-concise-4b-self/exp_002
+  mlflow_enabled: false
+  monitor_gpus: true
+  num_val_samples_to_print: 5
+  swanlab:
+    name: distillation-be-concise-OpenThoughtsMath-Qwen/Qwen3-4B-to-Qwen/Qwen3-4B-reverse-topk512
+    project: nemo-distillation
+  swanlab_enabled: false
+  tensorboard_enabled: false
+  wandb:
+    name: distill-be-concise-4b-self
+    project: nemo-distillation
+  wandb_enabled: true
+loss_fn:
+  kl_type: reverse
+  mixed_kl_weight: 0.5
+  zero_outside_topk: false
+policy:
+  dtensor_cfg:
+    _v2: true
+    activation_checkpointing: true
+    context_parallel_size: 2
+    cpu_offload: true
+    custom_parallel_plan: null
+    enabled: true
+    sequence_parallel: false
+    tensor_parallel_size: 4
+  dynamic_batching:
+    enabled: true
+    logprob_mb_tokens: 32768
+    sequence_length_round: 64
+    train_mb_tokens: 32768
+  generation:
+    _pad_token_id: 151643
+    backend: vllm
+    colocated:
+      enabled: true
+      resources:
+        gpus_per_node: null
+        num_nodes: null
+    max_new_tokens: 16384
+    model_name: Qwen/Qwen3-4B
+    stop_strings: null
+    stop_token_ids:
+    - 151645
+    temperature: 1.0
+    top_k: null
+    top_p: 1.0
+    vllm_cfg:
+      async_engine: false
+      distributed_executor_backend: null
+      enforce_eager: false
+      expert_parallel_size: 1
+      gpu_memory_utilization: 0.6
+      hf_overrides: {}
+      kv_cache_dtype: auto
+      load_format: dummy
+      max_model_len: 33280
+      num_first_layers_in_bf16: 0
+      num_last_layers_in_bf16: 0
+      pipeline_parallel_size: 1
+      precision: bfloat16
+      skip_tokenizer_init: true
+      tensor_parallel_size: 4
+      use_deep_gemm: false
+  generation_batch_size: 64
+  logprob_batch_size: 2
+  logprob_chunk_size: null
+  make_sequence_length_divisible_by: 16
+  max_grad_norm: 1.0
+  max_total_sequence_length: 17408
+  megatron_cfg:
+    activation_checkpointing: false
+    apply_rope_fusion: true
+    bias_activation_fusion: true
+    context_parallel_size: 2
+    converter_type: Qwen3ForCausalLM
+    defer_fp32_logits: false
+    distributed_data_parallel_config:
+      data_parallel_sharding_strategy: optim_grads_params
+      grad_reduce_in_fp32: false
+      overlap_grad_reduce: true
+      overlap_param_gather: true
+      use_custom_fsdp: false
+    empty_unused_memory_level: 0
+    enabled: false
+    expert_model_parallel_size: 1
+    expert_tensor_parallel_size: 1
+    fp8_cfg:
+      enabled: false
+      fp8: e4m3
+      fp8_param: false
+      fp8_recipe: blockwise
+    freeze_moe_router: true
+    moe_enable_deepep: false
+    moe_per_layer_logging: false
+    moe_permute_fusion: false
+    moe_router_bias_update_rate: 0.0
+    moe_router_dtype: fp64
+    moe_router_load_balancing_type: none
+    moe_shared_expert_overlap: false
+    moe_token_dispatcher_type: alltoall
+    num_layers_in_first_pipeline_stage: null
+    num_layers_in_last_pipeline_stage: null
+    optimizer:
+      adam_beta1: 0.9
+      adam_beta2: 0.999
+      adam_eps: 1.0e-08
+      bf16: true
+      clip_grad: 1.0
+      fp16: false
+      lr: 2.00001e-05
+      min_lr: 2.0e-05
+      optimizer: adam
+      optimizer_cpu_offload: false
+      optimizer_offload_fraction: 0.0
+      params_dtype: float32
+      sgd_momentum: 0.9
+      use_distributed_optimizer: true
+      use_precision_aware_optimizer: true
+      weight_decay: 0.01
+    pipeline_dtype: bfloat16
+    pipeline_model_parallel_size: 2
+    scheduler:
+      end_weight_decay: 0.01
+      lr_decay_iters: 1000
+      lr_decay_style: constant
+      lr_warmup_init: 2.0e-06
+      lr_warmup_iters: 10
+      start_weight_decay: 0.01
+      weight_decay_incr_style: constant
+    sequence_parallel: false
+    tensor_model_parallel_size: 2
+  model_name: Qwen/Qwen3-4B
+  offload_optimizer_for_logprob: true
+  optimizer:
+    kwargs:
+      betas:
+      - 0.9
+      - 0.999
+      eps: 1.0e-08
+      foreach: false
+      fused: false
+      lr: 5.0e-06
+      weight_decay: 0.01
+    name: torch.optim.AdamW
+  precision: bfloat16
+  scheduler:
+  - kwargs:
+      end_factor: 1.0
+      start_factor: 0.1
+      total_iters: 10
+    name: torch.optim.lr_scheduler.LinearLR
+  - kwargs:
+      factor: 1.0
+      total_iters: 10000000000
+    name: torch.optim.lr_scheduler.ConstantLR
+  - milestones:
+    - 10
+  sequence_packing:
+    algorithm: modified_first_fit_decreasing
+    enabled: false
+    logprob_mb_tokens: 34816
+    sequence_length_round: 64
+    train_mb_tokens: 34816
+  tokenizer:
+    chat_template_kwargs: null
+    name: Qwen/Qwen3-4B
+  train_global_batch_size: 64
+  train_micro_batch_size: 2
+teacher:
+  dtensor_cfg:
+    _v2: true
+    activation_checkpointing: true
+    context_parallel_size: 1
+    cpu_offload: true
+    custom_parallel_plan: null
+    enabled: true
+    sequence_parallel: false
+    tensor_parallel_size: 4
+  dynamic_batching:
+    enabled: true
+    logprob_mb_tokens: 32768
+    sequence_length_round: 64
+  generation:
+    backend: vllm
+    colocated:
+      enabled: true
+      resources:
+        gpus_per_node: null
+        num_nodes: null
+    max_new_tokens: 16384
+    stop_strings: null
+    stop_token_ids: null
+    temperature: 1.0
+    top_k: null
+    top_p: 1.0
+    vllm_cfg:
+      async_engine: false
+      distributed_executor_backend: null
+      enforce_eager: false
+      expert_parallel_size: 1
+      gpu_memory_utilization: 0.6
+      kv_cache_dtype: auto
+      max_model_len: 33280
+      num_first_layers_in_bf16: 0
+      num_last_layers_in_bf16: 0
+      pipeline_parallel_size: 1
+      precision: bfloat16
+      tensor_parallel_size: 4
+      use_deep_gemm: false
+  generation_batch_size: 64
+  logprob_batch_size: 2
+  logprob_chunk_size: null
+  make_sequence_length_divisible_by: 8
+  max_grad_norm: 1.0
+  max_total_sequence_length: 17408
+  megatron_cfg:
+    activation_checkpointing: false
+    apply_rope_fusion: true
+    bias_activation_fusion: true
+    context_parallel_size: 2
+    converter_type: Qwen3ForCausalLM
+    defer_fp32_logits: false
+    distributed_data_parallel_config:
+      data_parallel_sharding_strategy: optim_grads_params
+      grad_reduce_in_fp32: false
+      overlap_grad_reduce: true
+      overlap_param_gather: true
+      use_custom_fsdp: false
+    empty_unused_memory_level: 0
+    enabled: false
+    expert_model_parallel_size: 1
+    expert_tensor_parallel_size: 1
+    fp8_cfg:
+      enabled: false
+      fp8: e4m3
+      fp8_param: false
+      fp8_recipe: blockwise
+    freeze_moe_router: true
+    moe_enable_deepep: false
+    moe_per_layer_logging: false
+    moe_permute_fusion: false
+    moe_router_bias_update_rate: 0.0
+    moe_router_dtype: fp64
+    moe_router_load_balancing_type: none
+    moe_shared_expert_overlap: false
+    moe_token_dispatcher_type: alltoall
+    num_layers_in_first_pipeline_stage: null
+    num_layers_in_last_pipeline_stage: null
+    optimizer:
+      adam_beta1: 0.9
+      adam_beta2: 0.999
+      adam_eps: 1.0e-08
+      bf16: true
+      clip_grad: 1.0
+      fp16: false
+      lr: 2.00001e-05
+      min_lr: 2.0e-05
+      optimizer: adam
+      optimizer_cpu_offload: false
+      optimizer_offload_fraction: 0.0
+      params_dtype: float32
+      sgd_momentum: 0.9
+      use_distributed_optimizer: true
+      use_precision_aware_optimizer: true
+      weight_decay: 0.01
+    pipeline_dtype: bfloat16
+    pipeline_model_parallel_size: 2
+    scheduler:
+      end_weight_decay: 0.01
+      lr_decay_iters: 1000
+      lr_decay_style: constant
+      lr_warmup_init: 2.0e-06
+      lr_warmup_iters: 10
+      start_weight_decay: 0.01
+      weight_decay_incr_style: constant
+    sequence_parallel: false
+    tensor_model_parallel_size: 2
+  model_name: Qwen/Qwen3-4B
+  offload_optimizer_for_logprob: true
+  optimizer:
+    kwargs:
+      betas:
+      - 0.9
+      - 0.999
+      eps: 1.0e-08
+      foreach: false
+      fused: false
+      lr: 5.0e-06
+      weight_decay: 0.01
+    name: torch.optim.AdamW
+  precision: bfloat16
+  scheduler:
+  - kwargs:
+      end_factor: 1.0
+      start_factor: 0.1
+      total_iters: 10
+    name: torch.optim.lr_scheduler.LinearLR
+  - kwargs:
+      factor: 1.0
+      total_iters: 10000000000
+    name: torch.optim.lr_scheduler.ConstantLR
+  - milestones:
+    - 10
+  sequence_packing:
+    algorithm: modified_first_fit_decreasing
+    enabled: false
+    logprob_mb_tokens: 34816
+    sequence_length_round: 64
+    train_mb_tokens: 34816
+  tokenizer:
+    chat_template_kwargs: null
+    name: Qwen/Qwen3-4B
+  train_global_batch_size: 64
+  train_micro_batch_size: 2

step_25/policy/optimizer/optim/.metadata

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a646a64ca221a38d19777e0804a4800d94973aeef3554940be8a0635ec8a510c
+size 3744919

step_25/policy/optimizer/optim/__0_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:33b463be65569076cce859dcab0f8f63f05c5bc8e1ca79f312bf77bb9e9936cd
+size 2011935539

step_25/policy/optimizer/optim/__10_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b16e6f59fbf653269cb5fbec5ecb2e09ab1e22cf4d9d421c95bf448cbf8dc349
+size 2012773695

step_25/policy/optimizer/optim/__11_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1121359cf60e4a99010af3d3930ba8be5792d207cd5f42f4c7fb9ccc4c12bac5
+size 2012776378

step_25/policy/optimizer/optim/__12_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ae20123662ef473cf450ec1a965be302f0c8ed7c28917666ab807e8828619152
+size 2012489142

step_25/policy/optimizer/optim/__13_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:018e3e7f59dc79d1075e33f09340af98f1b1c5af7f119fda9641f906a79ec3e4
+size 2012766875

step_25/policy/optimizer/optim/__14_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f7b420b7bf1ce13a2648088904d8ad2ca30bfd33ff4c6715e360361cf4d5532e
+size 2012777162

step_25/policy/optimizer/optim/__15_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3a2eea58542b731f573dc4ccf1da3fa6da96d16c3fb59eec29a762a3479eada8
+size 2012779845

step_25/policy/optimizer/optim/__1_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:23b106800dd3529a9c8e30242f3dea10275a891bcf771439324ea15736d4e44f
+size 2012213272

step_25/policy/optimizer/optim/__2_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b41e447b536f76cc376d27febe8c92888706f92c3d6eb883a9d788095ea54eb6
+size 2012220092

step_25/policy/optimizer/optim/__3_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d19dc60d16d2a537e66ff46cf1a2c258026db5f279adbb92f1939083831c666d
+size 2012222775

step_25/policy/optimizer/optim/__4_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d965cadf3fec0d6a8c88a488ed411e70b7ae7748a6409aa4bc5b4058ab263c58
+size 2012489142

step_25/policy/optimizer/optim/__5_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d1c07160cf266c39fe03d2b458b0709b4e82e33f87a9e412e7bbf01f23e34b16
+size 2012766875

step_25/policy/optimizer/optim/__6_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0cfe2ff63199b567a38d1341784e6a396029c0122b3b0a2e0b8e9de846e4f363
+size 2012773695

step_25/policy/optimizer/optim/__7_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:af680ddbe668421eeb668e6b4f10f77fe6345276a9613b4b43c375d236980ec1
+size 2012776378

step_25/policy/optimizer/optim/__8_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7b7336d3f6c07ec8b338da0b78f925503edf653a4d077f8d078a804d0d2c26c
+size 2012489142

step_25/policy/optimizer/optim/__9_0.distcp

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:45beb5cb9e23ed77356fb5e9049f929ea2c2960697b8c586da59bcc067c3f827
+size 2012766875

step_25/policy/tokenizer/added_tokens.json

@@ -0,0 +1,28 @@
+{
+  "</think>": 151668,
+  "</tool_call>": 151658,
+  "</tool_response>": 151666,
+  "<think>": 151667,
+  "<tool_call>": 151657,
+  "<tool_response>": 151665,
+  "<|box_end|>": 151649,
+  "<|box_start|>": 151648,
+  "<|endoftext|>": 151643,
+  "<|file_sep|>": 151664,
+  "<|fim_middle|>": 151660,
+  "<|fim_pad|>": 151662,
+  "<|fim_prefix|>": 151659,
+  "<|fim_suffix|>": 151661,
+  "<|im_end|>": 151645,
+  "<|im_start|>": 151644,
+  "<|image_pad|>": 151655,
+  "<|object_ref_end|>": 151647,
+  "<|object_ref_start|>": 151646,
+  "<|quad_end|>": 151651,
+  "<|quad_start|>": 151650,
+  "<|repo_name|>": 151663,
+  "<|video_pad|>": 151656,
+  "<|vision_end|>": 151653,
+  "<|vision_pad|>": 151654,
+  "<|vision_start|>": 151652
+}

step_25/policy/tokenizer/chat_template.jinja

@@ -0,0 +1,89 @@
+{%- if tools %}
+    {{- '<|im_start|>system\n' }}
+    {%- if messages[0].role == 'system' %}
+        {{- messages[0].content + '\n\n' }}
+    {%- endif %}
+    {{- "# Tools\n\nYou may call one or more functions to assist with the user query.\n\nYou are provided with function signatures within <tools></tools> XML tags:\n<tools>" }}
+    {%- for tool in tools %}
+        {{- "\n" }}
+        {{- tool | tojson }}
+    {%- endfor %}
+    {{- "\n</tools>\n\nFor each function call, return a json object with function name and arguments within <tool_call></tool_call> XML tags:\n<tool_call>\n{\"name\": <function-name>, \"arguments\": <args-json-object>}\n</tool_call><|im_end|>\n" }}
+{%- else %}
+    {%- if messages[0].role == 'system' %}
+        {{- '<|im_start|>system\n' + messages[0].content + '<|im_end|>\n' }}
+    {%- endif %}
+{%- endif %}
+{%- set ns = namespace(multi_step_tool=true, last_query_index=messages|length - 1) %}
+{%- for message in messages[::-1] %}
+    {%- set index = (messages|length - 1) - loop.index0 %}
+    {%- if ns.multi_step_tool and message.role == "user" and message.content is string and not(message.content.startswith('<tool_response>') and message.content.endswith('</tool_response>')) %}
+        {%- set ns.multi_step_tool = false %}
+        {%- set ns.last_query_index = index %}
+    {%- endif %}
+{%- endfor %}
+{%- for message in messages %}
+    {%- if message.content is string %}
+        {%- set content = message.content %}
+    {%- else %}
+        {%- set content = '' %}
+    {%- endif %}
+    {%- if (message.role == "user") or (message.role == "system" and not loop.first) %}
+        {{- '<|im_start|>' + message.role + '\n' + content + '<|im_end|>' + '\n' }}
+    {%- elif message.role == "assistant" %}
+        {%- set reasoning_content = '' %}
+        {%- if message.reasoning_content is string %}
+            {%- set reasoning_content = message.reasoning_content %}
+        {%- else %}
+            {%- if '</think>' in content %}
+                {%- set reasoning_content = content.split('</think>')[0].rstrip('\n').split('<think>')[-1].lstrip('\n') %}
+                {%- set content = content.split('</think>')[-1].lstrip('\n') %}
+            {%- endif %}
+        {%- endif %}
+        {%- if loop.index0 > ns.last_query_index %}
+            {%- if loop.last or (not loop.last and reasoning_content) %}
+                {{- '<|im_start|>' + message.role + '\n<think>\n' + reasoning_content.strip('\n') + '\n</think>\n\n' + content.lstrip('\n') }}
+            {%- else %}
+                {{- '<|im_start|>' + message.role + '\n' + content }}
+            {%- endif %}
+        {%- else %}
+            {{- '<|im_start|>' + message.role + '\n' + content }}
+        {%- endif %}
+        {%- if message.tool_calls %}
+            {%- for tool_call in message.tool_calls %}
+                {%- if (loop.first and content) or (not loop.first) %}
+                    {{- '\n' }}
+                {%- endif %}
+                {%- if tool_call.function %}
+                    {%- set tool_call = tool_call.function %}
+                {%- endif %}
+                {{- '<tool_call>\n{"name": "' }}
+                {{- tool_call.name }}
+                {{- '", "arguments": ' }}
+                {%- if tool_call.arguments is string %}
+                    {{- tool_call.arguments }}
+                {%- else %}
+                    {{- tool_call.arguments | tojson }}
+                {%- endif %}
+                {{- '}\n</tool_call>' }}
+            {%- endfor %}
+        {%- endif %}
+        {{- '<|im_end|>\n' }}
+    {%- elif message.role == "tool" %}
+        {%- if loop.first or (messages[loop.index0 - 1].role != "tool") %}
+            {{- '<|im_start|>user' }}
+        {%- endif %}
+        {{- '\n<tool_response>\n' }}
+        {{- content }}
+        {{- '\n</tool_response>' }}
+        {%- if loop.last or (messages[loop.index0 + 1].role != "tool") %}
+            {{- '<|im_end|>\n' }}
+        {%- endif %}
+    {%- endif %}
+{%- endfor %}
+{%- if add_generation_prompt %}
+    {{- '<|im_start|>assistant\n' }}
+    {%- if enable_thinking is defined and enable_thinking is false %}
+        {{- '<think>\n\n</think>\n\n' }}
+    {%- endif %}
+{%- endif %}

step_25/policy/tokenizer/special_tokens_map.json

@@ -0,0 +1,31 @@
+{
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "eos_token": {
+    "content": "<|im_end|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  },
+  "pad_token": {
+    "content": "<|endoftext|>",
+    "lstrip": false,
+    "normalized": false,
+    "rstrip": false,
+    "single_word": false
+  }
+}

step_25/policy/tokenizer/tokenizer.json

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:aeb13307a71acd8fe81861d94ad54ab689df773318809eed3cbe794b4492dae4
+size 11422654

step_25/policy/tokenizer/tokenizer_config.json

@@ -0,0 +1,239 @@
+{
+  "add_bos_token": false,
+  "add_prefix_space": false,
+  "added_tokens_decoder": {
+    "151643": {
+      "content": "<|endoftext|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151644": {
+      "content": "<|im_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151645": {
+      "content": "<|im_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151646": {
+      "content": "<|object_ref_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151647": {
+      "content": "<|object_ref_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151648": {
+      "content": "<|box_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151649": {
+      "content": "<|box_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151650": {
+      "content": "<|quad_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151651": {
+      "content": "<|quad_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151652": {
+      "content": "<|vision_start|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151653": {
+      "content": "<|vision_end|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151654": {
+      "content": "<|vision_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151655": {
+      "content": "<|image_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151656": {
+      "content": "<|video_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": true
+    },
+    "151657": {
+      "content": "<tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151658": {
+      "content": "</tool_call>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151659": {
+      "content": "<|fim_prefix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151660": {
+      "content": "<|fim_middle|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151661": {
+      "content": "<|fim_suffix|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151662": {
+      "content": "<|fim_pad|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151663": {
+      "content": "<|repo_name|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151664": {
+      "content": "<|file_sep|>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151665": {
+      "content": "<tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151666": {
+      "content": "</tool_response>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151667": {
+      "content": "<think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    },
+    "151668": {
+      "content": "</think>",
+      "lstrip": false,
+      "normalized": false,
+      "rstrip": false,
+      "single_word": false,
+      "special": false
+    }
+  },
+  "additional_special_tokens": [
+    "<|im_start|>",
+    "<|im_end|>",
+    "<|object_ref_start|>",
+    "<|object_ref_end|>",
+    "<|box_start|>",
+    "<|box_end|>",
+    "<|quad_start|>",
+    "<|quad_end|>",
+    "<|vision_start|>",
+    "<|vision_end|>",
+    "<|vision_pad|>",
+    "<|image_pad|>",
+    "<|video_pad|>"
+  ],
+  "bos_token": null,
+  "clean_up_tokenization_spaces": false,
+  "eos_token": "<|im_end|>",
+  "errors": "replace",
+  "extra_special_tokens": {},
+  "model_max_length": 131072,
+  "pad_token": "<|endoftext|>",
+  "split_special_tokens": false,
+  "tokenizer_class": "Qwen2Tokenizer",
+  "unk_token": null
+}

step_25/policy/weights/model/.hf_metadata/config.json

@@ -0,0 +1,70 @@
+{
+  "architectures": [
+    "Qwen3ForCausalLM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "bos_token_id": 151643,
+  "dtype": "float32",
+  "eos_token_id": 151645,
+  "head_dim": 128,
+  "hidden_act": "silu",
+  "hidden_size": 2560,
+  "initializer_range": 0.02,
+  "intermediate_size": 9728,
+  "layer_types": [
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention",
+    "full_attention"
+  ],
+  "max_position_embeddings": 40960,
+  "max_window_layers": 36,
+  "model_type": "qwen3",
+  "nemo_version": "0.2.0",
+  "num_attention_heads": 32,
+  "num_hidden_layers": 36,
+  "num_key_value_heads": 8,
+  "pad_token_id": 151643,
+  "rms_norm_eps": 1e-06,
+  "rope_scaling": null,
+  "rope_theta": 1000000,
+  "sliding_window": null,
+  "tie_word_embeddings": true,
+  "transformers_version": "4.57.1",
+  "use_cache": false,
+  "use_sliding_window": false,
+  "vocab_size": 151936
+}

step_25/policy/weights/model/.hf_metadata/fqn_to_file_index_mapping.json

@@ -0,0 +1,400 @@
+{
+  "model.embed_tokens.weight": 1,
+  "model.layers.0.input_layernorm.weight": 1,
+  "model.layers.0.mlp.down_proj.weight": 1,
+  "model.layers.0.mlp.gate_proj.weight": 1,
+  "model.layers.0.mlp.up_proj.weight": 1,
+  "model.layers.0.post_attention_layernorm.weight": 1,
+  "model.layers.0.self_attn.k_norm.weight": 1,
+  "model.layers.0.self_attn.k_proj.weight": 1,
+  "model.layers.0.self_attn.o_proj.weight": 1,
+  "model.layers.0.self_attn.q_norm.weight": 1,
+  "model.layers.0.self_attn.q_proj.weight": 1,
+  "model.layers.0.self_attn.v_proj.weight": 1,
+  "model.layers.1.input_layernorm.weight": 1,
+  "model.layers.1.mlp.down_proj.weight": 1,
+  "model.layers.1.mlp.gate_proj.weight": 1,
+  "model.layers.1.mlp.up_proj.weight": 1,
+  "model.layers.1.post_attention_layernorm.weight": 1,
+  "model.layers.1.self_attn.k_norm.weight": 1,
+  "model.layers.1.self_attn.k_proj.weight": 1,
+  "model.layers.1.self_attn.o_proj.weight": 1,
+  "model.layers.1.self_attn.q_norm.weight": 1,
+  "model.layers.1.self_attn.q_proj.weight": 1,
+  "model.layers.1.self_attn.v_proj.weight": 1,
+  "model.layers.10.input_layernorm.weight": 1,
+  "model.layers.10.mlp.down_proj.weight": 1,
+  "model.layers.10.mlp.gate_proj.weight": 1,
+  "model.layers.10.mlp.up_proj.weight": 1,
+  "model.layers.10.post_attention_layernorm.weight": 1,
+  "model.layers.10.self_attn.k_norm.weight": 1,
+  "model.layers.10.self_attn.k_proj.weight": 1,
+  "model.layers.10.self_attn.o_proj.weight": 1,
+  "model.layers.10.self_attn.q_norm.weight": 1,
+  "model.layers.10.self_attn.q_proj.weight": 1,
+  "model.layers.10.self_attn.v_proj.weight": 1,
+  "model.layers.11.input_layernorm.weight": 1,
+  "model.layers.11.mlp.down_proj.weight": 1,
+  "model.layers.11.mlp.gate_proj.weight": 1,
+  "model.layers.11.mlp.up_proj.weight": 1,
+  "model.layers.11.post_attention_layernorm.weight": 1,
+  "model.layers.11.self_attn.k_norm.weight": 1,
+  "model.layers.11.self_attn.k_proj.weight": 1,
+  "model.layers.11.self_attn.o_proj.weight": 1,
+  "model.layers.11.self_attn.q_norm.weight": 1,
+  "model.layers.11.self_attn.q_proj.weight": 1,
+  "model.layers.11.self_attn.v_proj.weight": 1,
+  "model.layers.12.input_layernorm.weight": 1,
+  "model.layers.12.mlp.down_proj.weight": 1,
+  "model.layers.12.mlp.gate_proj.weight": 1,
+  "model.layers.12.mlp.up_proj.weight": 1,
+  "model.layers.12.post_attention_layernorm.weight": 1,
+  "model.layers.12.self_attn.k_norm.weight": 1,
+  "model.layers.12.self_attn.k_proj.weight": 1,
+  "model.layers.12.self_attn.o_proj.weight": 1,
+  "model.layers.12.self_attn.q_norm.weight": 1,
+  "model.layers.12.self_attn.q_proj.weight": 1,
+  "model.layers.12.self_attn.v_proj.weight": 1,
+  "model.layers.13.input_layernorm.weight": 1,
+  "model.layers.13.mlp.down_proj.weight": 1,
+  "model.layers.13.mlp.gate_proj.weight": 1,
+  "model.layers.13.mlp.up_proj.weight": 1,
+  "model.layers.13.post_attention_layernorm.weight": 1,
+  "model.layers.13.self_attn.k_norm.weight": 1,
+  "model.layers.13.self_attn.k_proj.weight": 1,
+  "model.layers.13.self_attn.o_proj.weight": 1,
+  "model.layers.13.self_attn.q_norm.weight": 1,
+  "model.layers.13.self_attn.q_proj.weight": 1,
+  "model.layers.13.self_attn.v_proj.weight": 1,
+  "model.layers.14.input_layernorm.weight": 1,
+  "model.layers.14.mlp.down_proj.weight": 1,
+  "model.layers.14.mlp.gate_proj.weight": 1,
+  "model.layers.14.mlp.up_proj.weight": 1,
+  "model.layers.14.post_attention_layernorm.weight": 1,
+  "model.layers.14.self_attn.k_norm.weight": 1,
+  "model.layers.14.self_attn.k_proj.weight": 1,
+  "model.layers.14.self_attn.o_proj.weight": 1,
+  "model.layers.14.self_attn.q_norm.weight": 1,
+  "model.layers.14.self_attn.q_proj.weight": 1,
+  "model.layers.14.self_attn.v_proj.weight": 1,
+  "model.layers.15.input_layernorm.weight": 1,
+  "model.layers.15.mlp.down_proj.weight": 1,
+  "model.layers.15.mlp.gate_proj.weight": 1,
+  "model.layers.15.mlp.up_proj.weight": 1,
+  "model.layers.15.post_attention_layernorm.weight": 1,
+  "model.layers.15.self_attn.k_norm.weight": 1,
+  "model.layers.15.self_attn.k_proj.weight": 1,
+  "model.layers.15.self_attn.o_proj.weight": 1,
+  "model.layers.15.self_attn.q_norm.weight": 1,
+  "model.layers.15.self_attn.q_proj.weight": 1,
+  "model.layers.15.self_attn.v_proj.weight": 1,
+  "model.layers.16.input_layernorm.weight": 1,
+  "model.layers.16.mlp.down_proj.weight": 1,
+  "model.layers.16.mlp.gate_proj.weight": 1,
+  "model.layers.16.mlp.up_proj.weight": 1,
+  "model.layers.16.post_attention_layernorm.weight": 1,
+  "model.layers.16.self_attn.k_norm.weight": 1,
+  "model.layers.16.self_attn.k_proj.weight": 1,
+  "model.layers.16.self_attn.o_proj.weight": 1,
+  "model.layers.16.self_attn.q_norm.weight": 1,
+  "model.layers.16.self_attn.q_proj.weight": 1,
+  "model.layers.16.self_attn.v_proj.weight": 1,
+  "model.layers.17.input_layernorm.weight": 1,
+  "model.layers.17.mlp.down_proj.weight": 1,
+  "model.layers.17.mlp.gate_proj.weight": 1,
+  "model.layers.17.mlp.up_proj.weight": 1,
+  "model.layers.17.post_attention_layernorm.weight": 1,
+  "model.layers.17.self_attn.k_norm.weight": 1,
+  "model.layers.17.self_attn.k_proj.weight": 1,
+  "model.layers.17.self_attn.o_proj.weight": 1,
+  "model.layers.17.self_attn.q_norm.weight": 1,
+  "model.layers.17.self_attn.q_proj.weight": 1,
+  "model.layers.17.self_attn.v_proj.weight": 1,
+  "model.layers.18.input_layernorm.weight": 1,
+  "model.layers.18.mlp.down_proj.weight": 1,
+  "model.layers.18.mlp.gate_proj.weight": 1,
+  "model.layers.18.mlp.up_proj.weight": 1,
+  "model.layers.18.post_attention_layernorm.weight": 1,
+  "model.layers.18.self_attn.k_norm.weight": 1,
+  "model.layers.18.self_attn.k_proj.weight": 1,
+  "model.layers.18.self_attn.o_proj.weight": 1,
+  "model.layers.18.self_attn.q_norm.weight": 1,
+  "model.layers.18.self_attn.q_proj.weight": 1,
+  "model.layers.18.self_attn.v_proj.weight": 1,
+  "model.layers.19.input_layernorm.weight": 1,
+  "model.layers.19.mlp.down_proj.weight": 1,
+  "model.layers.19.mlp.gate_proj.weight": 1,
+  "model.layers.19.mlp.up_proj.weight": 1,
+  "model.layers.19.post_attention_layernorm.weight": 1,
+  "model.layers.19.self_attn.k_norm.weight": 1,
+  "model.layers.19.self_attn.k_proj.weight": 1,
+  "model.layers.19.self_attn.o_proj.weight": 1,
+  "model.layers.19.self_attn.q_norm.weight": 1,
+  "model.layers.19.self_attn.q_proj.weight": 1,
+  "model.layers.19.self_attn.v_proj.weight": 1,
+  "model.layers.2.input_layernorm.weight": 1,
+  "model.layers.2.mlp.down_proj.weight": 1,
+  "model.layers.2.mlp.gate_proj.weight": 1,
+  "model.layers.2.mlp.up_proj.weight": 1,
+  "model.layers.2.post_attention_layernorm.weight": 1,
+  "model.layers.2.self_attn.k_norm.weight": 1,
+  "model.layers.2.self_attn.k_proj.weight": 1,
+  "model.layers.2.self_attn.o_proj.weight": 1,
+  "model.layers.2.self_attn.q_norm.weight": 1,
+  "model.layers.2.self_attn.q_proj.weight": 1,
+  "model.layers.2.self_attn.v_proj.weight": 1,
+  "model.layers.20.input_layernorm.weight": 1,
+  "model.layers.20.mlp.down_proj.weight": 1,
+  "model.layers.20.mlp.gate_proj.weight": 1,
+  "model.layers.20.mlp.up_proj.weight": 1,
+  "model.layers.20.post_attention_layernorm.weight": 1,
+  "model.layers.20.self_attn.k_norm.weight": 1,
+  "model.layers.20.self_attn.k_proj.weight": 1,
+  "model.layers.20.self_attn.o_proj.weight": 1,
+  "model.layers.20.self_attn.q_norm.weight": 1,
+  "model.layers.20.self_attn.q_proj.weight": 1,
+  "model.layers.20.self_attn.v_proj.weight": 1,
+  "model.layers.21.input_layernorm.weight": 1,
+  "model.layers.21.mlp.down_proj.weight": 1,
+  "model.layers.21.mlp.gate_proj.weight": 1,
+  "model.layers.21.mlp.up_proj.weight": 1,
+  "model.layers.21.post_attention_layernorm.weight": 1,
+  "model.layers.21.self_attn.k_norm.weight": 1,
+  "model.layers.21.self_attn.k_proj.weight": 1,
+  "model.layers.21.self_attn.o_proj.weight": 1,
+  "model.layers.21.self_attn.q_norm.weight": 1,
+  "model.layers.21.self_attn.q_proj.weight": 1,
+  "model.layers.21.self_attn.v_proj.weight": 1,
+  "model.layers.22.input_layernorm.weight": 1,
+  "model.layers.22.mlp.down_proj.weight": 1,
+  "model.layers.22.mlp.gate_proj.weight": 1,
+  "model.layers.22.mlp.up_proj.weight": 1,
+  "model.layers.22.post_attention_layernorm.weight": 1,
+  "model.layers.22.self_attn.k_norm.weight": 1,
+  "model.layers.22.self_attn.k_proj.weight": 1,
+  "model.layers.22.self_attn.o_proj.weight": 1,
+  "model.layers.22.self_attn.q_norm.weight": 1,
+  "model.layers.22.self_attn.q_proj.weight": 1,
+  "model.layers.22.self_attn.v_proj.weight": 1,
+  "model.layers.23.input_layernorm.weight": 1,
+  "model.layers.23.mlp.down_proj.weight": 1,
+  "model.layers.23.mlp.gate_proj.weight": 1,
+  "model.layers.23.mlp.up_proj.weight": 1,
+  "model.layers.23.post_attention_layernorm.weight": 1,
+  "model.layers.23.self_attn.k_norm.weight": 1,
+  "model.layers.23.self_attn.k_proj.weight": 1,
+  "model.layers.23.self_attn.o_proj.weight": 1,
+  "model.layers.23.self_attn.q_norm.weight": 1,
+  "model.layers.23.self_attn.q_proj.weight": 1,
+  "model.layers.23.self_attn.v_proj.weight": 1,
+  "model.layers.24.input_layernorm.weight": 1,
+  "model.layers.24.mlp.down_proj.weight": 1,
+  "model.layers.24.mlp.gate_proj.weight": 1,
+  "model.layers.24.mlp.up_proj.weight": 1,
+  "model.layers.24.post_attention_layernorm.weight": 1,
+  "model.layers.24.self_attn.k_norm.weight": 1,
+  "model.layers.24.self_attn.k_proj.weight": 1,
+  "model.layers.24.self_attn.o_proj.weight": 1,
+  "model.layers.24.self_attn.q_norm.weight": 1,
+  "model.layers.24.self_attn.q_proj.weight": 1,
+  "model.layers.24.self_attn.v_proj.weight": 1,
+  "model.layers.25.input_layernorm.weight": 1,
+  "model.layers.25.mlp.down_proj.weight": 1,
+  "model.layers.25.mlp.gate_proj.weight": 1,
+  "model.layers.25.mlp.up_proj.weight": 1,
+  "model.layers.25.post_attention_layernorm.weight": 1,
+  "model.layers.25.self_attn.k_norm.weight": 1,
+  "model.layers.25.self_attn.k_proj.weight": 1,
+  "model.layers.25.self_attn.o_proj.weight": 1,
+  "model.layers.25.self_attn.q_norm.weight": 1,
+  "model.layers.25.self_attn.q_proj.weight": 1,
+  "model.layers.25.self_attn.v_proj.weight": 1,
+  "model.layers.26.input_layernorm.weight": 1,
+  "model.layers.26.mlp.down_proj.weight": 1,
+  "model.layers.26.mlp.gate_proj.weight": 1,
+  "model.layers.26.mlp.up_proj.weight": 1,
+  "model.layers.26.post_attention_layernorm.weight": 1,
+  "model.layers.26.self_attn.k_norm.weight": 1,
+  "model.layers.26.self_attn.k_proj.weight": 1,
+  "model.layers.26.self_attn.o_proj.weight": 1,
+  "model.layers.26.self_attn.q_norm.weight": 1,
+  "model.layers.26.self_attn.q_proj.weight": 1,
+  "model.layers.26.self_attn.v_proj.weight": 1,
+  "model.layers.27.input_layernorm.weight": 1,
+  "model.layers.27.mlp.down_proj.weight": 1,
+  "model.layers.27.mlp.gate_proj.weight": 1,
+  "model.layers.27.mlp.up_proj.weight": 1,
+  "model.layers.27.post_attention_layernorm.weight": 1,
+  "model.layers.27.self_attn.k_norm.weight": 1,
+  "model.layers.27.self_attn.k_proj.weight": 1,
+  "model.layers.27.self_attn.o_proj.weight": 1,
+  "model.layers.27.self_attn.q_norm.weight": 1,
+  "model.layers.27.self_attn.q_proj.weight": 1,
+  "model.layers.27.self_attn.v_proj.weight": 1,
+  "model.layers.28.input_layernorm.weight": 1,
+  "model.layers.28.mlp.down_proj.weight": 1,
+  "model.layers.28.mlp.gate_proj.weight": 1,
+  "model.layers.28.mlp.up_proj.weight": 1,
+  "model.layers.28.post_attention_layernorm.weight": 1,
+  "model.layers.28.self_attn.k_norm.weight": 1,
+  "model.layers.28.self_attn.k_proj.weight": 1,
+  "model.layers.28.self_attn.o_proj.weight": 1,
+  "model.layers.28.self_attn.q_norm.weight": 1,
+  "model.layers.28.self_attn.q_proj.weight": 1,
+  "model.layers.28.self_attn.v_proj.weight": 1,
+  "model.layers.29.input_layernorm.weight": 1,
+  "model.layers.29.mlp.down_proj.weight": 1,
+  "model.layers.29.mlp.gate_proj.weight": 1,
+  "model.layers.29.mlp.up_proj.weight": 1,
+  "model.layers.29.post_attention_layernorm.weight": 1,
+  "model.layers.29.self_attn.k_norm.weight": 1,
+  "model.layers.29.self_attn.k_proj.weight": 1,
+  "model.layers.29.self_attn.o_proj.weight": 1,
+  "model.layers.29.self_attn.q_norm.weight": 1,
+  "model.layers.29.self_attn.q_proj.weight": 1,
+  "model.layers.29.self_attn.v_proj.weight": 1,
+  "model.layers.3.input_layernorm.weight": 1,
+  "model.layers.3.mlp.down_proj.weight": 1,
+  "model.layers.3.mlp.gate_proj.weight": 1,
+  "model.layers.3.mlp.up_proj.weight": 1,
+  "model.layers.3.post_attention_layernorm.weight": 1,
+  "model.layers.3.self_attn.k_norm.weight": 1,
+  "model.layers.3.self_attn.k_proj.weight": 1,
+  "model.layers.3.self_attn.o_proj.weight": 1,
+  "model.layers.3.self_attn.q_norm.weight": 1,
+  "model.layers.3.self_attn.q_proj.weight": 1,
+  "model.layers.3.self_attn.v_proj.weight": 1,
+  "model.layers.30.input_layernorm.weight": 1,
+  "model.layers.30.mlp.down_proj.weight": 1,
+  "model.layers.30.mlp.gate_proj.weight": 1,
+  "model.layers.30.mlp.up_proj.weight": 1,
+  "model.layers.30.post_attention_layernorm.weight": 1,
+  "model.layers.30.self_attn.k_norm.weight": 1,
+  "model.layers.30.self_attn.k_proj.weight": 1,
+  "model.layers.30.self_attn.o_proj.weight": 1,
+  "model.layers.30.self_attn.q_norm.weight": 1,
+  "model.layers.30.self_attn.q_proj.weight": 1,
+  "model.layers.30.self_attn.v_proj.weight": 1,
+  "model.layers.31.input_layernorm.weight": 1,
+  "model.layers.31.mlp.down_proj.weight": 1,
+  "model.layers.31.mlp.gate_proj.weight": 1,
+  "model.layers.31.mlp.up_proj.weight": 1,
+  "model.layers.31.post_attention_layernorm.weight": 1,
+  "model.layers.31.self_attn.k_norm.weight": 1,
+  "model.layers.31.self_attn.k_proj.weight": 1,
+  "model.layers.31.self_attn.o_proj.weight": 1,
+  "model.layers.31.self_attn.q_norm.weight": 1,
+  "model.layers.31.self_attn.q_proj.weight": 1,
+  "model.layers.31.self_attn.v_proj.weight": 1,
+  "model.layers.32.input_layernorm.weight": 1,
+  "model.layers.32.mlp.down_proj.weight": 1,
+  "model.layers.32.mlp.gate_proj.weight": 1,
+  "model.layers.32.mlp.up_proj.weight": 1,
+  "model.layers.32.post_attention_layernorm.weight": 1,
+  "model.layers.32.self_attn.k_norm.weight": 1,
+  "model.layers.32.self_attn.k_proj.weight": 1,
+  "model.layers.32.self_attn.o_proj.weight": 1,
+  "model.layers.32.self_attn.q_norm.weight": 1,
+  "model.layers.32.self_attn.q_proj.weight": 1,
+  "model.layers.32.self_attn.v_proj.weight": 1,
+  "model.layers.33.input_layernorm.weight": 1,
+  "model.layers.33.mlp.down_proj.weight": 1,
+  "model.layers.33.mlp.gate_proj.weight": 1,
+  "model.layers.33.mlp.up_proj.weight": 1,
+  "model.layers.33.post_attention_layernorm.weight": 1,
+  "model.layers.33.self_attn.k_norm.weight": 1,
+  "model.layers.33.self_attn.k_proj.weight": 1,
+  "model.layers.33.self_attn.o_proj.weight": 1,
+  "model.layers.33.self_attn.q_norm.weight": 1,
+  "model.layers.33.self_attn.q_proj.weight": 1,
+  "model.layers.33.self_attn.v_proj.weight": 1,
+  "model.layers.34.input_layernorm.weight": 1,
+  "model.layers.34.mlp.down_proj.weight": 1,
+  "model.layers.34.mlp.gate_proj.weight": 1,
+  "model.layers.34.mlp.up_proj.weight": 1,
+  "model.layers.34.post_attention_layernorm.weight": 1,
+  "model.layers.34.self_attn.k_norm.weight": 1,
+  "model.layers.34.self_attn.k_proj.weight": 1,
+  "model.layers.34.self_attn.o_proj.weight": 1,
+  "model.layers.34.self_attn.q_norm.weight": 1,
+  "model.layers.34.self_attn.q_proj.weight": 1,
+  "model.layers.34.self_attn.v_proj.weight": 1,
+  "model.layers.35.input_layernorm.weight": 1,
+  "model.layers.35.mlp.down_proj.weight": 1,
+  "model.layers.35.mlp.gate_proj.weight": 1,
+  "model.layers.35.mlp.up_proj.weight": 1,
+  "model.layers.35.post_attention_layernorm.weight": 1,
+  "model.layers.35.self_attn.k_norm.weight": 1,
+  "model.layers.35.self_attn.k_proj.weight": 1,
+  "model.layers.35.self_attn.o_proj.weight": 1,
+  "model.layers.35.self_attn.q_norm.weight": 1,
+  "model.layers.35.self_attn.q_proj.weight": 1,
+  "model.layers.35.self_attn.v_proj.weight": 1,
+  "model.layers.4.input_layernorm.weight": 1,
+  "model.layers.4.mlp.down_proj.weight": 1,
+  "model.layers.4.mlp.gate_proj.weight": 1,
+  "model.layers.4.mlp.up_proj.weight": 1,
+  "model.layers.4.post_attention_layernorm.weight": 1,
+  "model.layers.4.self_attn.k_norm.weight": 1,
+  "model.layers.4.self_attn.k_proj.weight": 1,
+  "model.layers.4.self_attn.o_proj.weight": 1,
+  "model.layers.4.self_attn.q_norm.weight": 1,
+  "model.layers.4.self_attn.q_proj.weight": 1,
+  "model.layers.4.self_attn.v_proj.weight": 1,
+  "model.layers.5.input_layernorm.weight": 1,
+  "model.layers.5.mlp.down_proj.weight": 1,
+  "model.layers.5.mlp.gate_proj.weight": 1,
+  "model.layers.5.mlp.up_proj.weight": 1,
+  "model.layers.5.post_attention_layernorm.weight": 1,
+  "model.layers.5.self_attn.k_norm.weight": 1,
+  "model.layers.5.self_attn.k_proj.weight": 1,
+  "model.layers.5.self_attn.o_proj.weight": 1,
+  "model.layers.5.self_attn.q_norm.weight": 1,
+  "model.layers.5.self_attn.q_proj.weight": 1,
+  "model.layers.5.self_attn.v_proj.weight": 1,
+  "model.layers.6.input_layernorm.weight": 1,
+  "model.layers.6.mlp.down_proj.weight": 1,
+  "model.layers.6.mlp.gate_proj.weight": 1,
+  "model.layers.6.mlp.up_proj.weight": 1,
+  "model.layers.6.post_attention_layernorm.weight": 1,
+  "model.layers.6.self_attn.k_norm.weight": 1,
+  "model.layers.6.self_attn.k_proj.weight": 1,
+  "model.layers.6.self_attn.o_proj.weight": 1,
+  "model.layers.6.self_attn.q_norm.weight": 1,
+  "model.layers.6.self_attn.q_proj.weight": 1,
+  "model.layers.6.self_attn.v_proj.weight": 1,
+  "model.layers.7.input_layernorm.weight": 1,
+  "model.layers.7.mlp.down_proj.weight": 1,
+  "model.layers.7.mlp.gate_proj.weight": 1,
+  "model.layers.7.mlp.up_proj.weight": 1,
+  "model.layers.7.post_attention_layernorm.weight": 1,
+  "model.layers.7.self_attn.k_norm.weight": 1,
+  "model.layers.7.self_attn.k_proj.weight": 1,
+  "model.layers.7.self_attn.o_proj.weight": 1,
+  "model.layers.7.self_attn.q_norm.weight": 1,
+  "model.layers.7.self_attn.q_proj.weight": 1,
+  "model.layers.7.self_attn.v_proj.weight": 1,
+  "model.layers.8.input_layernorm.weight": 1,
+  "model.layers.8.mlp.down_proj.weight": 1,
+  "model.layers.8.mlp.gate_proj.weight": 1,
+  "model.layers.8.mlp.up_proj.weight": 1,
+  "model.layers.8.post_attention_layernorm.weight": 1,
+  "model.layers.8.self_attn.k_norm.weight": 1,
+  "model.layers.8.self_attn.k_proj.weight": 1,
+  "model.layers.8.self_attn.o_proj.weight": 1,
+  "model.layers.8.self_attn.q_norm.weight": 1,
+  "model.layers.8.self_attn.q_proj.weight": 1,
+  "model.layers.8.self_attn.v_proj.weight": 1,
+  "model.layers.9.input_layernorm.weight": 1,
+  "model.layers.9.mlp.down_proj.weight": 1,
+  "model.layers.9.mlp.gate_proj.weight": 1,
+  "model.layers.9.mlp.up_proj.weight": 1,
+  "model.layers.9.post_attention_layernorm.weight": 1,
+  "model.layers.9.self_attn.k_norm.weight": 1,
+  "model.layers.9.self_attn.k_proj.weight": 1,
+  "model.layers.9.self_attn.o_proj.weight": 1,
+  "model.layers.9.self_attn.q_norm.weight": 1,
+  "model.layers.9.self_attn.q_proj.weight": 1,
+  "model.layers.9.self_attn.v_proj.weight": 1,
+  "model.norm.weight": 1
+}

step_25/policy/weights/model/.hf_metadata/generation_config.json

@@ -0,0 +1,13 @@
+{
+  "bos_token_id": 151643,
+  "do_sample": true,
+  "eos_token_id": [
+    151645,
+    151643
+  ],
+  "pad_token_id": 151643,
+  "temperature": 0.6,
+  "top_k": 20,
+  "top_p": 0.95,
+  "transformers_version": "4.57.1"
+}

step_25/policy/weights/model/shard-00001-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5bec0989ea98347e9c07fbb7998bce4063f551677d9e3727b7df17a45f24ecfd
+size 1005614616

step_25/policy/weights/model/shard-00002-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:d0d1820299e43d8abd174f56bfa957e1ed5d7fffb306bcf13eb3a9b91179f979
+size 1005688880

step_25/policy/weights/model/shard-00003-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e81a5247973d1ed30b217052715322e519e52b712f318042d032608dbfb7f1a
+size 1005688880

step_25/policy/weights/model/shard-00004-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f3ece813503c1a3c7324f146359f91b32859f2c8ec30efa7e43878a1f8e2db3c
+size 1005690656

step_25/policy/weights/model/shard-00005-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a44d0cc55d9252954998174b66b8410e9e0b2220f3872fde259de9056e59fef5
+size 1005615048

step_25/policy/weights/model/shard-00006-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8fc3c216ff514760c1fa7997b984dff76cd4912b93533cf92b6715daa92af86
+size 1005689096

step_25/policy/weights/model/shard-00007-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cfb5c91a3f22435766b27b5a3368aa28b21fc9778dcaf625a062181b96934688
+size 1005689096

step_25/policy/weights/model/shard-00008-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:82dd165c315bb9fdc5cbba16155e4461b59c6ccd3dd48395cadc886a7f25d2f4
+size 1005690872

step_25/policy/weights/model/shard-00009-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e3143156c0fd22967047a57d8a8e0888c4c7861932196c3956c2d7a78ae1732d
+size 1005615264

step_25/policy/weights/model/shard-00010-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a7474c0fd7e7e867c65caa2b9cb66b94f1416a327450f67a4e7e55bead0c40b8
+size 1005689192

step_25/policy/weights/model/shard-00011-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b4253d69924deb43245102e20ce8e9bd1aae9ced651247656cf299801d3597c
+size 1005689192

step_25/policy/weights/model/shard-00012-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:129ad8b62698ea3a9440f3a735b465f6f07641f457a3ad1539fd9e80fc731068
+size 1005690968

step_25/policy/weights/model/shard-00013-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9c0d05074935fc3f2fb66e761cd4788902bf8cc30700209e83a524e3d756cc77
+size 1005615264

step_25/policy/weights/model/shard-00014-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fd4fd1dcb4a7b48bb2a9e2deee2ff81ed6397eefec86a54011f7973d74be73f5
+size 1005689192

step_25/policy/weights/model/shard-00015-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efc28d8afa39f57e085e90df9d4e789856d56f1519a1df78af6a24ebf4955f28
+size 1005689192

step_25/policy/weights/model/shard-00016-model-00001-of-00001.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ce684938e09ec124c7e3ba161f3c8919367d8710e24be91cc09c2b9c0bbe9f8d
+size 1005690968

step_25/train_dataloader.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:69621e83bf24b13c2104e8c10f910b66a8705894bcddb770b99c0b016807fe6a
+size 7016

step_25/training_info.json

@@ -0,0 +1 @@
+{"current_epoch": 0, "current_step": 25, "total_steps": 25, "val_reward": 0.9182203389830509, "consumed_samples": 1600, "total_valid_tokens": 10104027.0, "val:accuracy": 0.9182203389830509}