1. Added memory management

2. Limited input text length
3. Limited response length
4. Added batch size control
5. Added thread limiting

src/app.py

@@ -206,7 +206,12 @@ llm = LlamaCPP(
     model_path=model_path,
     temperature=0.7,
     max_new_tokens=256,
-    context_window=2048
+    context_window=2048,
+    n_batch=512,      # Added batch size limit
+    n_ctx=2048,       # Explicit context window
+    verbose=False,    # Reduce logging
+    n_threads=4,      # Limit threads
+    last_n_tokens_size=256  # Limit token history
 )
 print("LLM initialized successfully")
 
@@ -556,31 +561,39 @@ with gr.Blocks(
     )
     
     def process_text_input(text, history):
-        """Process text input and generate response."""
+        """Process text input with memory management."""
         if not text:
             return history
 
         try:
+            # Limit input length
+            if len(text) > 500:
+                text = text[:500] + "..."
+                
             # Process the symptoms
             diagnosis_query = f"""
             Based on these symptoms: '{text}'
             Provide relevant ICD-10 codes and diagnostic questions.
             Focus on clinical implications.
+            Limit response to 1000 characters.
             """
             response = symptom_index.as_query_engine().query(diagnosis_query)
+            
+            # Clean up memory
+            cleanup_memory()
 
-            # Format and return chat messages
             return history + [
                 {"role": "user", "content": text},
                 {"role": "assistant", "content": format_response_for_user({
                     "diagnoses": [],
                     "confidences": [],
-                    "follow_up": str(response)
+                    "follow_up": str(response)[:1000]  # Limit response length
                 })}
             ]
 
         except Exception as e:
             print(f"Text processing error: {str(e)}")
+            cleanup_memory()
             return history
 
     submit_btn.click(