🩺 MedGemma-270M

MedGemma-270M is a 270M-parameter Gemma 3 model fine-tuned with LoRA on the MIRIAD-4.4M medical Q&A dataset.
This model is designed for fast, domain-specialized inference on small GPUs and CPUs.

---

Model Details

• Base Model: google/gemma-3-270m
• Parameters: 270M
• Fine-tuning Method: LoRA (r=8, alpha=16, dropout=0.0)
• Framework: Unsloth for efficient training
• Dataset: miriad-4.4M
• Task: Medical question answering & clinical reasoning

---

Training Configuration

• Epochs: 1
• Max Steps: 600
• Batch Size: 1 (grad_acc=24)
• Max Seq Length: 384
• Optimizer: AdamW 8-bit
• Precision: float16 (fp16)

---

Usage

Inference (Transformers)

from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_id = "OmerShah/medgemma-270m"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto")

prompt = "What are the common symptoms of iron deficiency anemia?"
inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
outputs = model.generate(**inputs, max_new_tokens=128)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))