Update app.py

app.py

@@ -8,7 +8,7 @@ ChatGPT-Premium-like open-source Gradio app with:
  - Caching of processed images & embeddings
  - Simple in-process queue & streaming text output
  - Rate-limiting per-client (token-bucket)
- 
+ h
 NOTES:
  - Replace model IDs with ones that match your hardware/quotas.
  - For production, swap the in-process queue with Redis/Celery and use S3/MinIO for storage.
@@ -423,6 +423,9 @@ def _stream_llm_response_generator(prompt: str, client_id: str):
     # final append
     session["history"].append(("Assistant", partial))
 
+# ---------------------------
+# GRADIO UI
+# ---------------------------
 # ---------------------------
 # GRADIO UI
 # ---------------------------
@@ -435,47 +438,55 @@ with gr.Blocks(css="""
 
     with gr.Row():
         with gr.Column(scale=8, elem_classes="chat-column"):
-            chatbot = gr.Chatbot(label="Assistant", elem_id="chatbot", show_label=False).scale(height=600)
+            chatbot = gr.Chatbot(
+                label="Assistant",
+                elem_id="chatbot",
+                show_label=False,
+                type="messages",
+                height=600
+            )
             with gr.Row():
-                txt = gr.Textbox(label="Type a message...", placeholder="Ask a question or upload images", show_label=False)
+                txt = gr.Textbox(
+                    label="Type a message...",
+                    placeholder="Ask a question or upload images",
+                    show_label=False
+                )
                 submit = gr.Button("Send")
             with gr.Row():
-                img_in = gr.File(label="Upload images (multiple)", file_count="multiple", file_types=["image"])
+                img_in = gr.File(
+                    label="Upload images (multiple)",
+                    file_count="multiple",
+                    file_types=["image"]
+                )
                 clear_btn = gr.Button("New Chat")
-            client_id_state = gr.State(str(uuid.uuid4()))  # simple per-window client id for rate limiting
+            client_id_state = gr.State(str(uuid.uuid4()))  # simple per-window client id
 
+    # ---------------------------
+    # handle send
+    # ---------------------------
     def handle_send(message, client_state, files):
         client_id = client_state or str(uuid.uuid4())
-        # process_request yields a generator; Gradio supports returning generator -> progressive updates
-        # We return a generator that yields strings; then the front-end should append them to chat.
         gen = process_request(client_id, files, message, stream=True)
-        # We'll wrap generator so Gradio can consume it; we will return a tuple (new user msg textbox, new history)
-        # But Gradio expects the function to return: (textbox_clear, new_chat_history)
-        # We'll implement a simple approach: produce a list of chunks and finally return them as a single assistant message.
         collected = ""
-        try:
-            for part in gen:
-                collected = part  # partial buffer
-                # return immediate partial update to be appended in chat — in current Gradio versions returning generator directly is best
-                yield "", [( "User", message ), ("Assistant", collected )]
-        except Exception as e:
-            yield "", [( "User", message ), ("Assistant", f"Error generating: {e}" )]
-        # final update (guarantee)
-        yield "", [( "User", message ), ("Assistant", collected )]
+        for part in gen:
+            collected = part
+            # Return in new type="messages" format
+            yield "", [
+                {"role": "user", "content": message},
+                {"role": "assistant", "content": collected}
+            ]
+        # Final guarantee
+        yield "", [
+            {"role": "user", "content": message},
+            {"role": "assistant", "content": collected}
+        ]
 
     # Connect send button and textbox
     submit.click(handle_send, inputs=[txt, client_id_state, img_in], outputs=[txt, chatbot])
     txt.submit(handle_send, inputs=[txt, client_id_state, img_in], outputs=[txt, chatbot])
 
+    # Clear chat button
     def clear_chat():
         client_id_state.value = str(uuid.uuid4())
         return [], ""
-    clear_btn.click(lambda: ([], "" ), None, [chatbot, txt])
-
-# initialize heavy models in background to avoid blocking Gradio start
-def bg_init():
-    init_visual_model()
-    init_math_model()
-threading.Thread(target=bg_init, daemon=True).start()
-
-demo.launch(server_name="0.0.0.0", server_port=7860, share=False)
+    clear_btn.click(clear_chat, None, [chatbot, txt])