Medgemma 4b It French Medical Assistant

🚀 MedGemma 4B-IT French Medical Assistant

This model is a fine-tuned version of google/medgemma-4b-it, specifically optimized to provide short and accurate medical responses in French.

The model has been trained to give concise responses (maximum 2 - 3 sentences) while maintaining medical accuracy.

✨ Features

• Base Model: google/medgemma-4b-it
• Language: French 🇫🇷
• Specialization: Short and accurate medical responses
• Technique: Direct fine-tuning
• Type: Text Generation
• Parameters: ~4B

📦 Installation

pip install transformers torch accelerate bitsandbytes

💻 Usage Examples

Basic Usage

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

# Configuration
model_id = "Sadou/medgemma-4b-it-french-medical-assistant"
device = "cuda" if torch.cuda.is_available() else "cpu"

# Loading the fine-tuned model
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=BitsAndBytesConfig(load_in_4bit=True),
    device_map="auto",
    torch_dtype=torch.bfloat16,
)

tokenizer = AutoTokenizer.from_pretrained(model_id)

def generate_medical_response(question, max_length=500, temperature=0.2):
    messages = [
        {
            "role": "system",
            "content": "Tu es un assistant médical expert. Réponds toujours en français de manière claire et précise."
        },
        {
            "role": "user", 
            "content": question
        }
    ]
    
    inputs = tokenizer.apply_chat_template(
        messages,
        add_generation_prompt=True,
        tokenize=True,
        return_dict=True,
        return_tensors="pt",
    ).to(model.device)
    
    input_len = inputs["input_ids"].shape[-1]
    
    with torch.inference_mode():
        outputs = model.generate(
            **inputs,
            max_new_tokens=max_length,
            do_sample=True,
            temperature=temperature,
            pad_token_id=tokenizer.eos_token_id,
           
        )
    
    response = tokenizer.decode(outputs[0][input_len:], skip_special_tokens=True)
    return response.strip()

# Example usage
question = "Comment traiter une migraine ?"
response = generate_medical_response(question)
print(f"Question: {question}")
print(f"Réponse: {response}")

Advanced Usage

import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

# Simple loading
model_id = "Sadou/medgemma-4b-it-french-medical-assistant"
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.bfloat16)
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Usage
question = "Comment traiter une migraine ?"
messages = [
    {"role": "system", "content": "Tu es un assistant médical expert."},
    {"role": "user", "content": question}
]

inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors="pt")
outputs = model.generate(inputs, max_new_tokens=200, temperature=0.7)
response = tokenizer.decode(outputs[0][inputs.shape[-1]:], skip_special_tokens=True)
print(response)

📚 Documentation

Example 1: Treatment

Input: "Comment traiter une migraine ?"

Output: "Prenez un antalgique (paracétamol 1g ou ibuprofène 600mg) dès les premiers signes. Reposez-vous dans un endroit sombre et silencieux avec une compresse froide sur le front. Si migraines fréquentes (>4/mois), consultez pour un traitement préventif."

⚠️ Important Note

• Medical Use: This model is for general information only.
• Not a Replacement: It does not replace a professional medical consultation.
• Emergencies: In case of a medical emergency, contact emergency services immediately.

👨‍💻 Author

Sadou BARRY

📄 License

Apache 2.0

Model created to improve access to medical information in French 🇫🇷