Update app.py

app.py

@@ -99,16 +99,18 @@ async def respond(message, history, system_message, max_tokens, temperature, top
         yield f"⚠️ Medical system error: {str(e)}"
 
 # ------ Enhanced Gradio Interface ------
+# ------ Gradio Interface ------
 demo = gr.ChatInterface(
     respond,
     additional_inputs=[
         gr.Textbox(value="Medical diagnosis assistant", label="System Role"),
-        gr.Slider(512, 1024, value=512, step=128, label="Max Tokens"),
-        gr.Slider(0.1, 1.0, value=0.5, step=0.1, label="Temperature"),
-        gr.Slider(0.7, 1.0, value=0.9, step=0.05, label="Top-p")
-    ],
-    examples=[["Headache and fever"], ["Chest pain radiating to arm"]]
-).queue(concurrency_count=5)  # Improved throughput [1]
+        gr.Slider(512, 2048, value=512, step=128, label="Max Tokens"),
+        gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(0.5, 1.0, value=0.95, step=0.05, label="Top-p")
+    ]
+).queue(concurrency_limit=5)  # Updated concurrency handling
 
 if __name__ == "__main__":
-    demo.launch(server_port=7860, share=True)
+    demo.launch(max_threads=10)  # Add thread pool configuration
+
+