Qwen3-0.6B-Math-Expert

Training Procedure

1. Dataset Preparation
   • The unsloth/OpenMathReasoning-mini dataset was used.
   • Each example was formatted in Chain-of-Thought (CoT) style, pairing math problems with step-by-step intermediate reasoning.
2. Model Loading and Configuration
   • Qwen3 base model weights were loaded via the unsloth library in bf16 precision.
   • All layers were updated (full_finetuning=True) to adapt the model for mathematical reasoning.
3. Supervised Fine-Tuning
   • Leveraged the Hugging Face TRL library with the Supervised Fine-Tuning (SFT) approach.
   • The model was trained to generate both correct answers and corresponding reasoning chains.

Purpose and Outcome

• The model’s reasoning capacity for math problems was significantly improved through single-dataset, full fine-tuning in bf16 precision.
• Outputs include both intermediate reasoning steps and final solutions, providing transparent and interpretable results.

License

Support