Code Repository for ExpLang

This is the code repository of the paper ExpLang: Improved Exploration and Exploitation in LLM Reasoning with On-Policy Thinking Language Selection.

Data and Model

• The high-quality SFT dataset for language-selective thinking is available at here on HuggingFace;
• The trained model based on Qwen3-4B is available at here on HuggingFace.