should wo

app.py

@@ -4,6 +4,7 @@ import time
 import gradio as gr
 import requests
 from transformers import AutoProcessor, PaliGemmaForConditionalGeneration
+from peft import PeftModel
 from PIL import Image
 from io import BytesIO
 import logging
@@ -19,30 +20,46 @@ device = None
 model_loaded = False
 
 def load_model():
-    """Load the AI model exactly like in Colab"""
+    """Load the AI model with PEFT adapter (Colab style)"""
     global model, processor, device, model_loaded
     
-    logger.info("Loading AI model (Colab style)...")
+    logger.info("Loading AI model with PEFT adapter (Colab style)...")
     
-    # === Load AI Model === (exactly like Colab)
-    model_id = "mychen76/paligemma-3b-mix-448-med_30k-ct-brain"
+    # === Load AI Model === (base model + adapter)
+    base_model_id = "google/paligemma-3b-mix-448"
+    adapter_model_id = "mychen76/paligemma-3b-mix-448-med_30k-ct-brain"
     device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
     dtype = torch.float16 if torch.cuda.is_available() else torch.float32
     
     logger.info(f"Using device: {device}")
     logger.info(f"Using dtype: {dtype}")
     logger.info(f"CUDA available: {torch.cuda.is_available()}")
+    logger.info(f"Base model: {base_model_id}")
+    logger.info(f"Adapter model: {adapter_model_id}")
     
     try:
-        # Load exactly like Colab (no token, no trust_remote_code)
+        # Load processor from base model
         logger.info("Loading processor...")
-        processor = AutoProcessor.from_pretrained(model_id)
+        processor = AutoProcessor.from_pretrained(base_model_id)
         
-        logger.info("Loading model...")
+        # Load base model
+        logger.info("Loading base model...")
         model = PaliGemmaForConditionalGeneration.from_pretrained(
-            model_id, 
-            torch_dtype=dtype
-        ).to(device).eval()
+            base_model_id, 
+            torch_dtype=dtype,
+            device_map="auto" if torch.cuda.is_available() else None
+        )
+        
+        # Load PEFT adapter
+        logger.info("Loading PEFT adapter...")
+        model = PeftModel.from_pretrained(model, adapter_model_id)
+        
+        # Set to eval mode
+        model.eval()
+        
+        # Move to device if not using device_map
+        if not torch.cuda.is_available():
+            model = model.to(device)
         
         logger.info("Model loaded successfully!")
         model_loaded = True
@@ -51,6 +68,15 @@ def load_model():
     except Exception as e:
         logger.error(f"Error loading model: {e}")
         logger.error(f"Error type: {type(e)}")
+        
+        # If license error, provide helpful message
+        if "license" in str(e).lower() or "access" in str(e).lower():
+            logger.error("This appears to be a license/access issue with the base model.")
+            logger.error("You may need to:")
+            logger.error("1. Accept the license for google/paligemma-3b-mix-448 on HuggingFace")
+            logger.error("2. Login with: huggingface-cli login")
+            logger.error("3. Use your HuggingFace token")
+        
         model_loaded = False
         return False
 
@@ -70,11 +96,17 @@ def analyze_brain_scan(image, patient_name="", patient_age="", symptoms=""):
 ## ⚠️ Model Loading Error
 
 The AI model is not available. This could be due to:
-- Model loading issues
-- Memory limitations
-- Network connectivity
+- **License Issue**: The base model requires accepting Google's license
+- **PEFT Loading Issue**: Problem loading the medical adapter
+- **Memory limitations**: Insufficient resources
+- **Network connectivity**: Download issues
+
+**To fix this:**
+1. Accept the license for `google/paligemma-3b-mix-448` on HuggingFace
+2. Login with your HuggingFace token: `huggingface-cli login`
+3. Restart the application
 
-Please check the logs or try refreshing.
+Please check the logs for more details.
 """
         
         if image is None:
@@ -102,6 +134,11 @@ Please check the logs or try refreshing.
 **AI Findings:**
 {result}
 
+**Model Info:**
+- Base Model: google/paligemma-3b-mix-448
+- Medical Adapter: mychen76/paligemma-3b-mix-448-med_30k-ct-brain
+- Device: {device}
+
 **Note:** This is an AI-generated analysis for educational purposes only. 
 Always consult with qualified medical professionals for actual diagnosis.
 """
@@ -116,7 +153,7 @@ def create_api_response(image, patient_name="", patient_age="", symptoms=""):
     """Create API-compatible response for integration"""
     try:
         if not model_loaded or model is None:
-            return {"error": "Model not loaded"}
+            return {"error": "Model not loaded - check license and authentication"}
         
         if image is None:
             return {"error": "No image provided"}
@@ -138,7 +175,8 @@ def create_api_response(image, patient_name="", patient_age="", symptoms=""):
                 "symptoms": symptoms
             },
             "model_info": {
-                "model_id": "mychen76/paligemma-3b-mix-448-med_30k-ct-brain",
+                "base_model": "google/paligemma-3b-mix-448",
+                "adapter_model": "mychen76/paligemma-3b-mix-448-med_30k-ct-brain",
                 "device": str(device),
                 "model_loaded": model_loaded
             }
@@ -161,10 +199,22 @@ def get_model_status():
 - **Model Object**: {type(model).__name__ if model else 'None'}
 - **Processor Object**: {type(processor).__name__ if processor else 'None'}
 - **PyTorch Version**: {torch.__version__}
+
+## 📋 Model Configuration
+
+- **Base Model**: google/paligemma-3b-mix-448
+- **Medical Adapter**: mychen76/paligemma-3b-mix-448-med_30k-ct-brain
+- **Model Type**: PEFT/LoRA Fine-tuned
+
+## ⚠️ Requirements
+
+- HuggingFace account with accepted license for PaliGemma
+- HuggingFace token authentication
+- PEFT library for adapter loading
 """
 
 # Load model at startup
-logger.info("Initializing Brain CT Analyzer (Colab Style)...")
+logger.info("Initializing Brain CT Analyzer with PEFT (Colab Style)...")
 load_success = load_model()
 if load_success:
     logger.info("Model loaded successfully!")
@@ -177,9 +227,11 @@ with gr.Blocks(title="Brain CT Analyzer", theme=gr.themes.Soft()) as demo:
     # 🧠 Brain CT Analyzer
     
     Upload a brain CT scan image for AI-powered analysis. This tool uses the PaliGemma medical model 
-    to provide preliminary findings.
+    with specialized medical fine-tuning to provide preliminary findings.
     
     **⚠️ Important:** This is for educational/research purposes only. Always consult qualified medical professionals.
+    
+    **🔑 Requirements:** This model requires accepting Google's PaliGemma license and HuggingFace authentication.
     """)
     
     # Model status section
@@ -221,7 +273,7 @@ with gr.Blocks(title="Brain CT Analyzer", theme=gr.themes.Soft()) as demo:
         with gr.Column(scale=1):
             result_output = gr.Markdown(
                 label="Analysis Results",
-                value="Upload an image and click 'Analyze Brain Scan' to see results." if model_loaded else "⚠️ Model not loaded. Check status above."
+                value="Upload an image and click 'Analyze Brain Scan' to see results." if model_loaded else "⚠️ Model not loaded. Check status above and ensure license acceptance."
             )
     
     # API endpoint simulation
@@ -245,16 +297,19 @@ with gr.Blocks(title="Brain CT Analyzer", theme=gr.themes.Soft()) as demo:
     # Instructions
     gr.Markdown("""
     ## 📋 Usage Instructions:
-    1. Upload a brain CT scan image (JPEG or PNG)
-    2. Optionally fill in patient information
-    3. Click "Analyze Brain Scan" to get AI findings
-    4. Review the results in the output panel
+    1. **Accept License**: Go to [google/paligemma-3b-mix-448](https://huggingface.co/google/paligemma-3b-mix-448) and accept the license
+    2. **Authenticate**: Login with `huggingface-cli login` using your token
+    3. Upload a brain CT scan image (JPEG or PNG)
+    4. Optionally fill in patient information
+    5. Click "Analyze Brain Scan" to get AI findings
+    6. Review the results in the output panel
     
     ## 🔗 Integration:
     This interface can be integrated with your medical app using the Gradio API.
     
     ## ✅ Based on Working Colab Code:
-    This version uses the exact same model loading and inference code as your working Google Colab notebook.
+    This version uses PEFT to load the medical fine-tuned adapter on top of the base PaliGemma model,
+    exactly matching your working Google Colab setup.
     """)
 
 if __name__ == "__main__":
@@ -262,4 +317,4 @@ if __name__ == "__main__":
         server_name="0.0.0.0",
         server_port=7860,
         share=True
-    )
+    ) 

requirements.txt

@@ -1,5 +1,10 @@
 torch>=2.0.0
+torchvision>=0.15.0
 transformers>=4.40.0
+peft>=0.10.0
+accelerate>=0.20.0
 gradio>=4.0.0
-pillow>=9.0.0
-accelerate>=0.20.0 
+Pillow>=9.0.0
+requests>=2.25.0
+numpy>=1.21.0
+huggingface_hub>=0.16.0