Fine-tuning DeepSeek R1 with QLoRA.

This repository contains notebook to fine-tune DeepSeek R1 8B model using Unsloth. Unsloth allows for faster and more memory-efficient fine-tuning of LLMs.

Training Run Screenshots

Colab QLoRA Training Run

Weights & Biases QLoRA Training Run

Contents

The notebooks/ directory contains the following workflows:

• medical_finetuning_using_hf_unsloth.ipynb: Focuses on QLoRA (Quantized LoRA) techniques in 4-bit quantization for efficient fine-tuning.

Requirements

• Python 3.10+
• CUDA-enabled GPU (recommended for Unsloth)

Quick Setup

1. Install Jupyter VS Code extension (if not installed):
2. Launch Jupyter Lab or Notebook:

      jupyter lab

3. Open one of the notebooks in the notebooks/ directory and follow the steps.

• Prepare Secrets for notebook:

• HF_TOKEN – Required for HuggingFace login.

• WANDB_TOKEN – Required for Weights & Biases (W&B) logging. Optional if you do not want training charts.

• Execute all runs and wait until training is complete:

Contributing

Contributions welcome — open an issue or submit a PR.

License

See the LICENSE file for details.