local inference page with fixes

app.py

@@ -20,7 +20,10 @@ logger = logging.getLogger(__name__)
 APP_VERSION = "1.0.0"
 logger.info(f"Starting Radiology Teaching App v{APP_VERSION}")
 
-# Initialize models for local inference
+# Global variables
+pipe = None
+llm = None
+tokenizer = None
 device = 0 if torch.cuda.is_available() else "cpu"
 logger.info(f"Using device: {device}")
 
@@ -130,6 +133,8 @@ def analyze_with_llama(
     temperature: float = 0.6,
 ) -> Iterator[str]:
     """Analyze transcribed report against ground truth using Llama"""
+    global llm, tokenizer  # Add global declaration
+    
     if llm is None or tokenizer is None:
         raise gr.Error("Llama model not initialized properly!")
     
@@ -280,11 +285,12 @@ with gr.Blocks() as demo:
         )
         
         # Load case for local analysis
+        local_image_display = gr.Image(label="Chest X-ray Image", type="pil")  # Add this line
         load_case_btn.click(
             fn=load_random_case,
             inputs=[gr.Checkbox(value=False, visible=False)],  # Hidden checkbox for hide_ground_truth
             outputs=[
-                gr.Image(visible=False),  # Hidden image output
+                local_image_display,  # Update this line
                 local_ground_truth_findings,
                 local_ground_truth_impression,
                 gr.State(),  # Hidden state