Update app.py

app.py

@@ -1013,48 +1013,49 @@ with gr.Blocks(theme="soft") as demo:
         feedback_grp = gr.Group(visible=False)
         with feedback_grp:
             gr.Markdown("### How helpful was this answer?")
-            rating = gr.Radio(choices=[1,2,3,4,5], label="Rating (1=poor, 5=great)")
+            rating = gr.Radio(choices=[1, 2, 3, 4, 5], label="Rating (1=poor, 5=great)")
             comment = gr.Textbox(label="Optional comment", placeholder="What was good or missing?")
             submit_fb = gr.Button("Submit feedback")
             fb_status = gr.Markdown("")
 
-# Wiring
-enter_btn.click(
-    fn=enter_app,
-    inputs=[first_tb, last_tb, state],
-    outputs=[gate, app, state, gate_msg],
-)
+    # ---- Wiring (MUST stay inside the Blocks context) ----
+    enter_btn.click(
+        fn=enter_app,
+        inputs=[first_tb, last_tb, state],
+        outputs=[gate, app, state, gate_msg],
+    )
 
-send_btn.click(
-    fn=predict,
-    inputs=[user_in, chat, state],
-    outputs=[chat, user_in, feedback_grp, rating, comment, state],
-    concurrency_limit=1,  # serialize LLM calls
-)
+    send_btn.click(
+        fn=predict,
+        inputs=[user_in, chat, state],
+        outputs=[chat, user_in, feedback_grp, rating, comment, state],
+        concurrency_limit=1,  # serialize LLM calls
+    )
 
-user_in.submit(
-    fn=predict,
-    inputs=[user_in, chat, state],
-    outputs=[chat, user_in, feedback_grp, rating, comment, state],
-    concurrency_limit=1,  # serialize LLM calls
-)
+    user_in.submit(
+        fn=predict,
+        inputs=[user_in, chat, state],
+        outputs=[chat, user_in, feedback_grp, rating, comment, state],
+        concurrency_limit=1,  # serialize LLM calls
+    )
 
-clear_btn.click(
-    fn=lambda: ([], "", gr.update(visible=False), None, "", init_session()),
-    inputs=None,
-    outputs=[chat, user_in, feedback_grp, rating, comment, state],
-    concurrency_limit=4,
-)
+    clear_btn.click(
+        fn=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,
-)
+    submit_fb.click(
+        fn=save_feedback,
+        inputs=[rating, comment, state],
+        outputs=[fb_status, feedback_grp],
+        concurrency_limit=4,
+    )
 
-# Queue (true concurrency = 1 to prevent OOM/restarts)
+# Queue & launch (outside the Blocks)
 demo.queue(concurrency_count=1, max_size=64)
+demo.launch(max_threads=8)