Upload 3 files

Dockerfile

@@ -0,0 +1,25 @@
+FROM python:3.11-slim
+
+# Set working directory
+WORKDIR /app
+
+# Install system dependencies
+RUN apt-get update && apt-get install -y \
+    build-essential \
+    curl \
+    software-properties-common \
+    git \
+    && rm -rf /var/lib/apt/lists/*
+
+# Copy requirements and install
+COPY requirements.txt .
+RUN pip install --no-cache-dir -r requirements.txt
+
+# Copy the application code
+COPY . .
+
+# Expose port (HF Spaces uses 7860)
+EXPOSE 7860
+
+# Run the app
+CMD ["python", "app.py"]

app.py

@@ -0,0 +1,79 @@
+from fastapi import FastAPI, UploadFile, File, Form
+from transformers import pipeline
+from PIL import Image
+import torch
+import io
+import uvicorn
+import os
+
+app = FastAPI(title="MedGemma 4B Internal API")
+
+# Check if we are running on Hugging Face Spaces
+# Spaces usually provide GPUs, if not it will fallback to CPU (will be slow)
+device = "cuda" if torch.cuda.is_available() else "cpu"
+dtype = torch.bfloat16 if torch.cuda.is_available() else torch.float32
+
+print(f"Loading full 9GB MedGemma model on {device}...")
+
+try:
+    # Use the pipeline API for the easiest implementation of the 9GB model
+    pipe = pipeline(
+        "image-text-to-text",
+        model="google/medgemma-4b-it",
+        torch_dtype=dtype,
+        device_map="auto",
+    )
+    print("Model loaded successfully!")
+except Exception as e:
+    print(f"Error loading model: {e}")
+    pipe = None
+
+@app.get("/")
+def read_root():
+    return {
+        "status": "MedGemma 4B API is active",
+        "device": device,
+        "model_size": "Full 9GB"
+    }
+
+@app.post("/analyze")
+async def analyze_image(
+    prompt: str = Form("Describe this medical image and give a preliminary analysis."),
+    file: UploadFile = File(...)
+):
+    if pipe is None:
+        return {"error": "Model not loaded properly. Check logs."}
+
+    # Read the uploaded image
+    contents = await file.read()
+    image = Image.open(io.BytesIO(contents)).convert("RGB")
+    
+    # Format messages for MedGemma
+    messages = [
+        {
+            "role": "system",
+            "content": [{"type": "text", "text": "You are an expert radiologist and medical consultant."}]
+        },
+        {
+            "role": "user",
+            "content": [
+                {"type": "text", "text": prompt},
+                {"type": "image", "image": image}
+            ]
+        }
+    ]
+    
+    # Inference
+    output = pipe(text=messages, max_new_tokens=250)
+    
+    # Extract the response text
+    result = output[0]["generated_text"][-1]["content"]
+    
+    return {
+        "analysis": result,
+        "success": True
+    }
+
+if __name__ == "__main__":
+    # Port 7860 is the default for Hugging Face Spaces
+    uvicorn.run(app, host="0.0.0.0", port=7860)

requirements.txt

@@ -0,0 +1,8 @@
+transformers>=4.50.0
+accelerate
+torch
+pillow
+fastapi
+uvicorn
+python-multipart
+requests