Llama-3.2-3B-claude-3.7-sonnet-reasoning-distilled/README.md

base_model, tags, license, language, datasets

base_model	tags	license	language	datasets
unsloth/llama-3.2-3b-instruct	text-generation-inference transformers unsloth llama gguf	apache-2.0	en	mfielding92/claude-3.7-sonnet-reasoning

Uploaded Model

Overview

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.

Technical Details

• Developed by: mfielding92
• Base Model: unsloth/llama-3.2-3b-instruct
• Finetuning Method: Supervised Fine-Tuning (SFT) using LoRA
• Training Speed Enhancement: Trained 2x faster with Unsloth and Huggingface's TRL library

Training Data

The model was fine-tuned on a dataset derived from:

• mfielding92/claude-3.7-sonnet-reasoning

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.

Usage Notes

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.