Add 3B recipe

recipes/R1-Distill-Qwen-3B/sft/config_v00.00.yaml

@@ -0,0 +1,55 @@
+# Config for 2 nodes
+# Model arguments
+model_name_or_path: Qwen/Qwen2.5-3B
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/reasoning-mix
+dataset_config: all
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 4
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: true
+hub_model_id: open-r1/R1-Distill-Qwen-3B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-3B
+overwrite_output_dir: true
+per_device_eval_batch_size: 2
+per_device_train_batch_size: 4
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger_kernel: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-7B-Instruct/sft/config_v00.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: Qwen/Qwen2.5-7B-Instruct
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/reasoning-mix
+dataset_config: all
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-7B-Instruct
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-7B-Instruct
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-7B/sft/config_v00.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: Qwen/Qwen2.5-7B
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/OpenR1-Math-220k
+dataset_config: default
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-7B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-7B/sft/config_v01.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: Qwen/Qwen2.5-7B
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/codeforces-cots_decontaminated
+dataset_config: solutions_all_cpp_py
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-7B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-7B/sft/config_v02.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: Qwen/Qwen2.5-7B
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/Llama-Nemotron-Post-Training-Dataset-Processed
+dataset_config: science_r1
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-7B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-7B/sft/config_v03.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: Qwen/Qwen2.5-Math-7B
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/OpenR1-Math-220k
+dataset_config: default
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-7B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v00.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/OpenR1-Math-220k
+dataset_config: default
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v01.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/codeforces-cots_decontaminated
+dataset_config: solutions_all_cpp_py
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v01.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v02.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/Llama-Nemotron-Post-Training-Dataset-Processed
+dataset_config: science_r1
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v02.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v03.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/reasoning-mix
+dataset_config: all
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v03.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v04.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/OpenR1-Math-220k
+dataset_config: extended
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v04.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v05.00.yaml

@@ -0,0 +1,55 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/OpenR1-Math-220k
+dataset_config: all
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v04.00
+hub_strategy: every_save
+learning_rate: 1.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v06.00.yaml

@@ -0,0 +1,74 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_mixture:
+  datasets:                     
+    - id: open-r1/OpenR1-Math-220k   
+      config: all     
+      split: train            
+      columns:
+        - messages
+      weight: 1.0                                 # 225k samples
+    - id: open-r1/codeforces-cots_decontaminated
+      config: solutions_all_cpp_py
+      split: train         
+      columns:
+        - messages
+      weight: 1.0                                 # 83.1k samples
+    - id: open-r1/Llama-Nemotron-Post-Training-Dataset-Processed
+      config: science_r1
+      split: train         
+      columns:
+        - messages
+      weight: 1.0                                 # 173k samples
+  seed: 0
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v06.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v07.00.yaml

@@ -0,0 +1,68 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_mixture:
+  datasets:                     
+    - id: open-r1/OpenR1-Math-220k   
+      config: default     
+      split: train            
+      columns:
+        - messages
+      weight: 1.0                                 # 93.7k samples
+    - id: open-r1/codeforces-cots_decontaminated
+      config: solutions_all_cpp_py
+      split: train         
+      columns:
+        - messages
+      weight: 0.011                               # ~1% of total mix
+  seed: 0
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v07.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v08.00.yaml

@@ -0,0 +1,68 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_mixture:
+  datasets:                     
+    - id: open-r1/OpenR1-Math-220k   
+      config: default     
+      split: train            
+      columns:
+        - messages
+      weight: 1.0                                 # 93.7k samples
+    - id: open-r1/codeforces-cots_decontaminated
+      config: solutions_all_cpp_py
+      split: train         
+      columns:
+        - messages
+      weight: 0.056                               # ~5% of total mix
+  seed: 0
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v08.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v09.00.yaml

@@ -0,0 +1,68 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_mixture:
+  datasets:                     
+    - id: open-r1/OpenR1-Math-220k   
+      config: default     
+      split: train            
+      columns:
+        - messages
+      weight: 1.0                                 # 93.7k samples
+    - id: open-r1/codeforces-cots_decontaminated
+      config: solutions_all_cpp_py
+      split: train         
+      columns:
+        - messages
+      weight: 0.113                               # ~10% of total mix
+  seed: 0
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v09.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v10.00.yaml

@@ -0,0 +1,74 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_mixture:
+  datasets:                     
+    - id: open-r1/OpenR1-Math-220k   
+      config: default     
+      split: train            
+      columns:
+        - messages
+      weight: 1.0                                 # 93.7k samples
+    - id: open-r1/codeforces-cots_decontaminated
+      config: solutions_all_cpp_py
+      split: train         
+      columns:
+        - messages
+      weight: 0.06                                # ~5k samples (~5% of total mix)
+    - id: open-r1/Llama-Nemotron-Post-Training-Dataset-Processed
+      config: science_r1
+      split: train
+      columns:
+        - messages
+      weight: 0.029                               # ~5k samples (~5% of total mix) 
+  seed: 0
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v10.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v11.00.yaml

@@ -0,0 +1,74 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_mixture:
+  datasets:                     
+    - id: open-r1/OpenR1-Math-220k   
+      config: default     
+      split: train            
+      columns:
+        - messages
+      weight: 0.053                                 # ~5k samples (~5% of total mix)
+    - id: open-r1/codeforces-cots_decontaminated
+      config: solutions_all_cpp_py
+      split: train         
+      columns:
+        - messages
+      weight: 1.0                                   # 83.1k samples
+    - id: open-r1/Llama-Nemotron-Post-Training-Dataset-Processed
+      config: science_r1
+      split: train
+      columns:
+        - messages
+      weight: 0.029                                 # ~5k samples (~5% of total mix) 
+  seed: 0
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v11.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen-Math-7B/sft/config_v12.00.yaml

@@ -0,0 +1,74 @@
+# Config for 1 node
+# Model arguments
+model_name_or_path: open-r1/Qwen2.5-Math-7B-RoPE-300k
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_mixture:
+  datasets:                     
+    - id: open-r1/OpenR1-Math-220k   
+      config: default     
+      split: train            
+      columns:
+        - messages
+      weight: 0.106                                 # ~10k samples (~5% of total mix)
+    - id: open-r1/codeforces-cots_decontaminated
+      config: solutions_all_cpp_py
+      split: train         
+      columns:
+        - messages
+      weight: 0.12                                  # ~10k samples (~5% of total mix)
+    - id: open-r1/Llama-Nemotron-Post-Training-Dataset-Processed
+      config: science_r1
+      split: train
+      columns:
+        - messages
+      weight: 1.0                                   # ~173k samples
+  seed: 0
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 8
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen-Math-7B
+hub_model_revision: v12.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen-Math-7B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger: true
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen3-30B-A3B/sft/config_v00.00.yaml

@@ -0,0 +1,56 @@
+# Config for 8 nodes
+# Model arguments
+model_name_or_path: Qwen/Qwen3-30B-A3B-Base
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/reasoning-mix
+dataset_config: all
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+# activation_offloading: true
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 2
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: false
+hub_model_id: open-r1/R1-Distill-Qwen3-30B-A3B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen3-30B-A3B
+overwrite_output_dir: true
+per_device_eval_batch_size: 1
+per_device_train_batch_size: 1
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger_kernel: true # Cannot use at the same time as act offload
+warmup_ratio: 0.03

recipes/R1-Distill-Qwen3-8B/sft/config_v00.00.yaml

@@ -0,0 +1,55 @@
+# Config for 2 nodes
+# Model arguments
+model_name_or_path: Qwen/Qwen3-8B-Base
+model_revision: main
+torch_dtype: bfloat16
+attn_implementation: flash_attention_2
+
+# Data training arguments
+chat_template: "{%- if tools %}\n    {{- '<|im_start|>system\\n' }}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- messages[0]['content'] }}\n    {%- else %}\n        {{- 'You are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.' }}\n    {%- endif %}\n    {{- \"\\n\\n# 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>\" }}\n    {%- for tool in tools %}\n        {{- \"\\n\" }}\n        {{- tool | tojson }}\n    {%- endfor %}\n    {{- \"\\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\" }}\n{%- else %}\n    {%- if messages[0]['role'] == 'system' %}\n        {{- '<|im_start|>system\\n' + messages[0]['content'] + '<|im_end|>\\n' }}\n    {%- else %}\n        {{- '<|im_start|>system\\nYou are Open-R1, a language model trained by Hugging Face to help users. Your role as an assistant involves thoroughly exploring questions through a systematic thinking process before providing the final precise and accurate solutions. This requires engaging in a comprehensive cycle of analysis, summarizing, exploration, reassessment, reflection, backtracing, and iteration to develop well-considered thinking process. Please structure your response into two main sections: Thought and Solution using the specified format: <think> Thought section </think> Solution section. In the Thought section, detail your reasoning process in steps. Each step should include detailed considerations such as analysing questions, summarizing relevant findings, brainstorming new ideas, verifying the accuracy of the current steps, refining any errors, and revisiting previous steps. In the Solution section, based on various attempts, explorations, and reflections from the Thought section, systematically present the final solution that you deem correct. The Solution section should be logical, accurate, and concise and detail necessary steps needed to reach the conclusion. Now, try to solve the following question through the above guidelines.<|im_end|>\\n' }}\n    {%- endif %}\n{%- endif %}\n{%- for message in messages %}\n    {%- if (message.role == \"user\") or (message.role == \"system\" and not loop.first) or (message.role == \"assistant\" and not message.tool_calls) %}\n        {{- '<|im_start|>' + message.role + '\\n' + message.content + '<|im_end|>' + '\\n' }}\n    {%- elif message.role == \"assistant\" %}\n        {{- '<|im_start|>' + message.role }}\n        {%- if message.content %}\n            {{- '\\n' + message.content }}\n        {%- endif %}\n        {%- for tool_call in message.tool_calls %}\n            {%- if tool_call.function is defined %}\n                {%- set tool_call = tool_call.function %}\n            {%- endif %}\n            {{- '\\n<tool_call>\\n{\"name\": \"' }}\n            {{- tool_call.name }}\n            {{- '\", \"arguments\": ' }}\n            {{- tool_call.arguments | tojson }}\n            {{- '}\\n</tool_call>' }}\n        {%- endfor %}\n        {{- '<|im_end|>\\n' }}\n    {%- elif message.role == \"tool\" %}\n        {%- if (loop.index0 == 0) or (messages[loop.index0 - 1].role != \"tool\") %}\n            {{- '<|im_start|>user' }}\n        {%- endif %}\n        {{- '\\n<tool_response>\\n' }}\n        {{- message.content }}\n        {{- '\\n</tool_response>' }}\n        {%- if loop.last or (messages[loop.index0 + 1].role != \"tool\") %}\n            {{- '<|im_end|>\\n' }}\n        {%- endif %}\n    {%- endif %}\n{%- endfor %}\n{%- if add_generation_prompt %}\n    {{- '<|im_start|>assistant\\n' }}\n{%- endif %}\n"
+dataset_name: open-r1/reasoning-mix
+dataset_config: all
+dataset_num_proc: 12
+eos_token: <|im_end|>
+
+# SFT trainer config
+callbacks:
+- push_to_hub_revision
+benchmarks:
+- aime24
+- gpqa
+- lcb_v4
+bf16: true
+do_eval: false
+eval_strategy: 'no'
+gradient_accumulation_steps: 4
+gradient_checkpointing: true
+gradient_checkpointing_kwargs:
+  use_reentrant: true
+hub_model_id: open-r1/R1-Distill-Qwen3-8B
+hub_model_revision: v00.00
+hub_strategy: every_save
+learning_rate: 4.0e-05
+log_level: info
+logging_steps: 1
+logging_strategy: steps
+lr_scheduler_type: cosine_with_min_lr
+lr_scheduler_kwargs:
+  min_lr_rate: 0.1
+packing: false
+max_grad_norm: 0.2
+max_length: 32768
+max_steps: -1
+num_train_epochs: 5
+output_dir: data/R1-Distill-Qwen3-8B
+overwrite_output_dir: true
+per_device_eval_batch_size: 2
+per_device_train_batch_size: 2
+push_to_hub: true
+report_to:
+- wandb
+save_strategy: epoch
+save_total_limit: 1
+seed: 42
+use_liger_kernel: true
+warmup_ratio: 0.03