This model is a Llama 3.2 3B Instruct variant that has been specifically fine-tuned using reasoning data from Claude 3.7 Sonnet. The goal was to leverage Claude's renowned reasoning capabilities within a more accessible, open-source architecture like Llama.
This allows the model to potentially exhibit improved logical thinking, problem-solving abilities, and complex reasoning compared to the base Llama 3.2 model while remaining open-source.
While this model inherits some of Claude's strenghs in reasoning, it is still a derivative work built on Llama architecture. Users should evaluate its performance carefully for their specific use cases.
This llama model was trained 2x faster with
Unsloth and Huggingface's TRL library.
