Official implementation of DualDistill: A trajectory-composition distillation method for integrating tool use into long-chain-of-thought reasoning.

Weihua Du, Pranjal Aggarwal, Sean Welleck, & Yiming Yang
"Agentic-R1: Distilled Dual-Strategy Reasoning." (2025)

• Efficient Training: Integrates tool use into long-chain-of-thought (CoT) reasoning using only 4 × A6000 GPUs
• Unified Reasoning: Fuses heterogeneous reasoning traces from multiple teacher models into a single student model

Overview of DualDistill methodology

• Agentic-R1 demonstrates significant performance gains on DeepMath-L and Combinatorics300, where both complex reasoning and tool use are crucial for success.
• Agentic-R1-SD (Self-Distilled) further enhances performance through our self-distillation approach, consistently outperforming baseline models across nearly all evaluation tasks.

1. Clone the repository:

   git clone https://github.com/StigLidu/DualDistill.git
   cd DualDistill

2. Create environment (optional but recommended):

   conda create -n dualdistill python=3.11
   conda activate dualdistill

3. Install dependencies:

   pip install -r requirements.txt
   pip install flash-attn --no-build-isolation

Download the base model:

python script/data_script/model_download.py \
  --repo_id deepseek-ai/DeepSeek-R1-Distill-Qwen-7B \
  --local_dir models

Prepare training data:

python script/data_script/teacher_data_download.py

Train the student model using teacher trajectories:

bash script/sft_script/SFT.sh

Start inference server:

bash script/eval_script/start_inference_server.sh [model_path] [display_name] [port]

Sample self-distillation trajectories:

python sft/self_distillation_sampler.py \
  --server_url http://localhost:$port/v1 \
  --model_name [display_name] \
  --model_path [model_path] \
  --save_path [path_to_save_trajectories]

Prepare self-distillation data:

# Extract teacher solutions
python script/data_script/extract_training_solution.py

# Construct training dataset
python script/data_script/processing_self_distillation_traj.py

Fine-tune on self-distillation data:

bash script/sft_script/expert_iteration.sh [model_path] [data_path] [save_path]

bash script/eval_script/start_inference_server.sh [model_path] [display_name] [port]

bash script/eval_script/eval_remote_server.sh \
  [url] [display_name] [data_path] [code_mode] [max_token]

Example:

bash script/eval_script/eval_remote_server.sh \
  "http://localhost:8080/v1" "agentic-r1" "dataset/test/math.json" "true" "4096"

• Code Execution Safety: The evaluation scripts execute model-generated code locally. Only use trusted models before execution.
• Inference Config: If you are using vLLM (a recent version) and encounter an error regarding the maximum context length. You may need to modify the model_max_length in tokenizer_config.json.
• Self-Distillation Warning: The self-distillation step requires sampling many trajectories and can be time-consuming.

This project is licensed under the MIT License - see the LICENSE file for details.

We thank the following open-source projects for their foundational contributions:

• OpenHands - Agent framework
• DeepMath-103K - Mathematical reasoning dataset
• vLLM - High-performance inference engine

For questions or support, please contact:

• Weihua Du: [email protected]

If you find our work useful, please consider citing:

@article{du2025agentic,
  title={Agentic-R1: Distilled Dual-Strategy Reasoning},
  author={Du, Weihua and Aggarwal, Pranjal and Welleck, Sean and Yang, Yiming},
  journal={arXiv preprint arXiv:2507.05707},
  year={2025}
}