Update app.py

app.py

@@ -1,9 +1,8 @@
 import gradio as gr
 from transformers import pipeline
 
-# 1. Initialize the pipeline
-# device_map="auto" automatically loads the model onto the GPU if available
 print("Loading model...")
+# device_map="auto" automatically loads the model onto the GPU if available
 pipe = pipeline(
     "image-text-to-text",
     model="huihui-ai/Huihui-Qwen3.5-2B-abliterated",
@@ -12,7 +11,7 @@ pipe = pipeline(
 print("Model loaded successfully!")
 
 def respond(message, history):
-    # 2. Reconstruct the message history in the format expected by the model
+    # Reconstruct the message history in the format expected by the model
     messages =[]
     
     for user_msg, assistant_msg in history:
@@ -24,38 +23,35 @@ def respond(message, history):
         if assistant_msg:
             messages.append({
                 "role": "assistant",
-                "content": [{"type": "text", "text": assistant_msg}]
+                "content":[{"type": "text", "text": assistant_msg}]
             })
             
-    # 3. Add the current user message
+    # Add the current user message
     messages.append({
         "role": "user",
         "content": [{"type": "text", "text": message}]
     })
     
-    # 4. Generate the response
-    # Added max_new_tokens so the model's response doesn't cut off early
-    response = pipe(text=messages, max_new_tokens=1024)
+    # Generate the response
+    # FIX: max_new_tokens=2048 prevents the output from cutting off at 250 characters
+    response = pipe(text=messages, max_new_tokens=2048)
     
-    # 5. Extract the assistant's response dynamically
-    # The pipeline returns the full conversation history. 
-    # The newly generated response is ALWAYS the last item in the list [-1]
+    # Extract the assistant's response dynamically
     generated_messages = response[0]['generated_text']
     assistant_reply = generated_messages[-1]['content']
     
-    # 6. Return the text properly whether the model outputs a list of dicts or a raw string
+    # Return the text properly whether the model outputs a list of dicts or a raw string
     if isinstance(assistant_reply, list):
         reply_text = "".join([item.get("text", "") for item in assistant_reply if item.get("type") == "text"])
         return reply_text
     else:
         return assistant_reply
 
-# 7. Set up the Gradio Chat Interface
+# FIX: Removed `theme="soft"` which was causing the TypeError crash in Gradio 4
 demo = gr.ChatInterface(
     fn=respond,
     title="Huihui Qwen3.5 Chatbot",
     description="A conversational AI using the `huihui-ai/Huihui-Qwen3.5-2B-abliterated` model.",
-    theme="soft",
     examples=["How to make a book?", "Explain quantum physics to a child.", "Write a poem about AI."]
 )