Update app.py

app.py

@@ -79,15 +79,18 @@ def process_chat_streaming(user_message, chatbot_display, messages_list, image_p
             stop_strings=["<|endoftext|>", "User:"] # Add stop strings to prevent over-generation
         )
 
-        # Create generation kwargs for the thread
-        generation_kwargs = dict(
-            inputs,
-            streamer=streamer,
-            generation_config=generation_config
+        # *** THE FIX IS HERE ***
+        # We must pass 'inputs' as a positional argument for 'batch'
+        # and the rest as keyword arguments.
+        thread = Thread(
+            target=model.generate_from_batch,
+            args=[inputs],  # Pass `inputs` as the first positional argument ('batch')
+            kwargs={       # Pass the rest as keyword arguments
+                "generation_config": generation_config,
+                "tokenizer": processor.tokenizer,
+                "streamer": streamer,
+            }
         )
-
-        # Run generation in a separate thread
-        thread = Thread(target=model.generate_from_batch, kwargs=generation_kwargs)
         thread.start()
 
         # Yield updates to the Gradio UI
@@ -136,8 +139,6 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="neutra
                     scale=4,
                     container=False
                 )
-                # The submit button is now primarily for show; Enter key is the main way to submit
-                # but we will wire it up anyway.
 
     # --- Event Listeners ---
 
@@ -146,7 +147,6 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="neutra
         fn=process_chat_streaming,
         inputs=[user_textbox, chatbot_display, messages_list, image_input],
         outputs=[chatbot_display, messages_list],
-        # queue=False # Set queue to False for faster interaction with streaming
     )
 
     # Chain the action to also clear the textbox after submission
@@ -167,4 +167,4 @@ with gr.Blocks(theme=gr.themes.Default(primary_hue="blue", secondary_hue="neutra
 
 
 if __name__ == "__main__":
-    demo.launch(debug=True, mcp_server=True)
+    demo.launch(mcp_server=True)