Update app.py

app.py

@@ -906,23 +906,37 @@ with gr.Blocks(theme="soft") as demo:
         outputs=[gate, app, state, gate_msg],
     )
 
-    send_btn.click(predict, inputs=[user_in, chat, state],
-                   outputs=[chat, user_in, feedback_grp, rating, comment, state])
-    user_in.submit(predict, inputs=[user_in, chat, state],
-                   outputs=[chat, user_in, feedback_grp, rating, comment, state])
+    send_btn.click(
+        predict,
+        inputs=[user_in, chat, state],
+        outputs=[chat, user_in, feedback_grp, rating, comment, state],
+        concurrency_limit=8,  # heavy LLM call
+    )
+
+    user_in.submit(
+        predict,
+        inputs=[user_in, chat, state],
+        outputs=[chat, user_in, feedback_grp, rating, comment, state],
+        concurrency_limit=8,  # heavy LLM call
+    )
 
-    clear_btn.click(lambda: ([], "", gr.update(visible=False), None, "", init_session()),
-                    inputs=None,
-                    outputs=[chat, user_in, feedback_grp, rating, comment, state])
+    clear_btn.click(
+        lambda: ([], "", gr.update(visible=False), None, "", init_session()),
+        inputs=None,
+        outputs=[chat, user_in, feedback_grp, rating, comment, state],
+        concurrency_limit=4,
+    )
 
     submit_fb.click(
         fn=save_feedback,
         inputs=[rating, comment, state],
         outputs=[fb_status, feedback_grp],
+        concurrency_limit=4,
     )
 
-demo.queue(
-    concurrency_count=int(os.environ.get("CONCURRENCY", "2")),
-    max_size=64
-).launch()
+
+
+demo.queue(max_size=64)
+demo.launch(max_threads=int(os.environ.get("MAX_THREADS", "32")))
+