agnihotri-anxh/HealthMate-gemma-medical-lora

🩺 HealthMate – Gemma 2B Medical LoRA Model

HealthMate is a fine-tuned LoRA adapter built on Google Gemma-2B-IT, trained on medically-oriented text extracted from the Gale Encyclopedia of Medicine. The model specializes in encyclopedia-style medical Q&A, sticking strictly to source content and avoiding medical advice.

💡 Model Summary

• —Base Model: google/gemma-2b-it
• —Type: LoRA Adapter (PEFT)
• —Parameters Trained: ~20M (LoRA layers only)
• —Task: Medical question answering, explanation, terminology, definitions
• —Language: English
• —Format: Instruction → Input → Output
• —Strength: Produces concise, reference-style medical answers
• —Avoids: Personal medical advice, diagnosis, or clinical recommendations

🧠 Model Description

HealthMate was fine-tuned to mimic the structured, factual tone of medical encyclopedias. It is NOT a medical decision support tool. It is designed purely for educational, informational, and academic use.

✨ Output Characteristics

• —Always begins with “According to the Gale Encyclopedia of Medicine:”
• —Uses a neutral, encyclopedic writing style
• —Does not hallucinate clinical advice
• —Structured and consistent formatting
• —Follows your training template exactly

📘 Training Dataset

The model was trained on a custom dataset derived from:

• —Gale Encyclopedia of Medicine (text extracted using OCR + manual cleanup)
• —~20,000 instruction-style samples
• —Each sample contains:
• —instruction (task description)
• —input (question)
• —output (encyclopedia-derived answer)

🚀 How to Use

from transformers import AutoModelForCausalLM, AutoTokenizer
from peft import PeftModel

base = AutoModelForCausalLM.from_pretrained(
    "google/gemma-2b-it",
    load_in_4bit=True,
    device_map="auto"
)

model = PeftModel.from_pretrained(base, "agnihotri-anxh/HealthMate-gemma-medical-lora")
tokenizer = AutoTokenizer.from_pretrained("google/gemma-2b-it")

prompt = """Instruction: Answer the question based ONLY on the book content. Do NOT give medical advice.
Input: Explain ultrasound in simple terms.
Output:"""

inputs = tokenizer(prompt, return_tensors="pt").to("cuda")
outputs = model.generate(**inputs, max_new_tokens=200)
print(tokenizer.decode(outputs[0]))