This model is a Gemma 3 4B variant specifically fine-tuned using reasoning data from Claude 3.7 Sonnet. The goal was to integrate Claude's acclaimed reasoning capabilities within a powerful, open-source architecture like Gemma.
This enables the model to potentially demonstrate superior logical reasoning, complex problem-solving, and analytical thinking compared to the standard Gemma 3 model, while remaining accessible and open-source.
While this model incorporates some of Claude's reasoning strengths, it remains a derivative built on Gemma architecture. Users should thoroughly evaluate its performance for specific tasks and applications.
This Gemma model was trained 2x faster with
Unsloth and Huggingface's TRL library.
