license: apache-2.0
language:
- en
tags:
- medical
- gemma
- lora
- unsloth
- healthcare
pipeline_tag: text-generation
MedQA Gemma 1B LoRA
A medical question-answering LoRA fine-tuning project using Gemma 3 1B and Unsloth.
MedQA Gemma 1B LoRA
A medical question-answering LoRA fine-tuning project using Gemma 3 1B and Unsloth.
Model Overview
This model was fine-tuned on a subset of the MedQA-USMLE dataset using:
- Unsloth
- LoRA
- Google Colab
- Tesla T4 GPU
The goal of this project is to explore medical instruction tuning and lightweight domain adaptation for healthcare-related question answering.
Base Model
unsloth/gemma-3-1b-it
Dataset
GBaker/MedQA-USMLE-4-options
Training Details
- Training Method: LoRA
- GPU: Tesla T4
- Framework: Unsloth
- Max Steps: 30
- Training Samples: 500
- Precision: float32
Example Prompt
Question:
A pregnant woman has dysuria without fever or flank pain.
Options:
A. Doxycycline
B. Nitrofurantoin
C. Vancomycin
D. Oseltamivir
Example Output
B. Nitrofurantoin
Intended Use
This project is intended for:
- AI experimentation
- educational purposes
- medical LLM fine-tuning research
Limitations
This model is NOT intended for:
- real clinical diagnosis
- medical treatment decisions
- production healthcare deployment
It was trained on a limited subset for experimentation purposes only.
Tools Used
- Hugging Face Transformers
- Unsloth
- TRL
- PEFT
- LoRA
- Google Colab
Author
Ayman Mohammed
