Llama3.2:3B Fine-Tuning with Unsloth on BAAI/Infinity-Instruct

This repository demonstrates the fine-tuning of the Llama3.2:3B model using the BAAI/Infinity-Instruct dataset and the Unsloth library. The fine-tuned model, MateoRov/Llama3.2-3b-SFF-Infinity-MateoRovere, is now available on Hugging Face and can be used for conversational AI tasks.

---

Features

• Fine-tunes the Llama3.2:3B model for supervised instruction-following tasks.
• Utilizes Unsloth for efficient and scalable training.
• Leverages the BAAI/Infinity-Instruct dataset for high-quality supervised fine-tuning.
• Provides a terminal-based chat interface using the fine-tuned model.

---

Setup and Usage

1. Clone the Repository

git clone https://github.com/Mateorovere/FineTuning-LLM-Llama3.2-3b.git
cd FineTuning-LLM-Llama3.2-3b

2. Install Dependencies

Install the required Python packages:

pip install -r requirements.txt

And install PyTorch

3. Run Fine-Tuning (Optional)

If you want to perform fine-tuning yourself, open the Llama_3_2_3B_Finetuning.ipynb notebook and follow the steps to train the model using the Unsloth library and the BAAI/Infinity-Instruct dataset.

Both the Llama_3_2_3B_Finetuning.ipynb and the inference.ipynb are ment to run on google collab

4. Use the Fine-Tuned Model as a Chatbot

Run the main.py script to interact with the fine-tuned model in the terminal:

python main.py

Model Information

The fine-tuned model is hosted on Hugging Face:

• Model: MateoRov/Llama3.2-3b-SFF-Infinity-MateoRovere
• Dataset: BAAI/Infinity-Instruct

Contributing

Contributions are welcome! Please fork the repository and create a pull request with your proposed changes.

Future improvements

Making it compatible with Ollama