oldflag's picture
Update README.md
d5eb40b verified
metadata
language:
  - en
license: apache-2.0
tags:
  - medical
  - unsloth
datasets:
  - oldflag/symptom_dx_test
pipeline_tag: question-answering

Fine-Tuning Llama3-8b-bnb-4bit Model for Medical Symptom Diagnosis

This project demonstrates how to fine-tune the Llama3-8b-bnb-4bit model using a Question and Answer dataset focused on medical symptoms and their diagnoses.
The project is implemented using Google Colab and utilizes the unsloth library for efficient model handling.

Overview

The goal of this project is to fine-tune the Llama3-8b-bnb-4bit model to generate accurate medical diagnoses based on input symptoms.
This is achieved by using a dataset of medical Q&A pairs and adapting the model to understand and respond to medical queries effectively.

Setup and Installation

  1. Clone the repository and navigate to the project directory:

    git clone https://github.com/oldfalg/FineTuning_Llama_3_8b_Symptom_Dx.git
    cd FineTuning_Llama_3_8b_Symptom_Dx
    

Key Components

• Model Loading:
Utilizes the FastLanguageModel from the unsloth library to load the pre-trained Llama3-8b-bnb-4bit model with 4-bit quantization for efficient memory usage.
• Dataset Preparation:
Uses the datasets library to load and process a Q&A dataset for fine-tuning.
• Fine-Tuning:
Fine-tunes the model in Colab to generate accurate diagnoses based on input symptoms.
• Model Uploading:
Supports saving the fine-tuned model in different formats (float16, int4, and LoRA adapters) and uploading it to Hugging Face.

Inference

After fine-tuning, the model can be used to generate diagnoses based on new symptom inputs.
The project supports enabling native faster inference and using the fine-tuned model for generation tasks.