Update app.py

app.py

@@ -5,7 +5,7 @@ import torch
 
 # --- CONFIGURATION ---
 BASE_MODEL = "unsloth/Qwen3-VL-2B-Instruct-unsloth-bnb-4bit"
-LORA_ID = "EthanCastro/qwen3-vl-2b-quickdraw"  # <--- Checked: Matches your repo name
+LORA_ID = "EthanCastro/qwen3-vl-2b-quickdraw" 
 
 print("Loading model and processor...")
 model = AutoModelForImageTextToText.from_pretrained(
@@ -15,26 +15,38 @@ model = AutoModelForImageTextToText.from_pretrained(
     trust_remote_code=True
 )
 
-# Load your LoRA adapters
 model = PeftModel.from_pretrained(model, LORA_ID)
 processor = AutoProcessor.from_pretrained("Qwen/Qwen3-VL-2B-Instruct", trust_remote_code=True)
 print("Model Ready!")
 
 def respond(message, image, history):
+    # History is now a list of dictionaries
+    # Format: [{"role": "user", "content": "hi"}, {"role": "assistant", "content": "hello"}]
     messages = []
-    # Build conversation history
-    for user_msg, assistant_msg in history:
-        messages.append({"role": "user", "content": [{"type": "text", "text": user_msg}]})
-        messages.append({"role": "assistant", "content": [{"type": "text", "text": assistant_msg}]})
+    
+    # 1. Convert history to Qwen's multimodal format
+    for msg in history:
+        # We need to ensure content is treated as text for the history buffer
+        content = msg["content"]
+        # If content is a list (multimodal), extract just the text for simplicity
+        if isinstance(content, list):
+            text_content = next((item['text'] for item in content if item['type'] == 'text'), "")
+        else:
+            text_content = content
+            
+        messages.append({
+            "role": msg["role"], 
+            "content": [{"type": "text", "text": text_content}]
+        })
 
-    # Add current user turn
+    # 2. Add current user turn with the new image
     user_content = []
     if image is not None:
         user_content.append({"type": "image", "image": image})
     user_content.append({"type": "text", "text": message})
     messages.append({"role": "user", "content": user_content})
 
-    # Tokenize and Generate
+    # 3. Tokenize and Generate
     text = processor.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
     
     if image is not None:
@@ -47,7 +59,6 @@ def respond(message, image, history):
     
     generated_text = processor.batch_decode(outputs, skip_special_tokens=True)[0]
     
-    # Extract only the assistant's response
     if "assistant" in generated_text:
         response = generated_text.split("assistant")[-1].strip()
     else:
@@ -56,11 +67,12 @@ def respond(message, image, history):
     return response
 
 # --- GRADIO INTERFACE ---
-with gr.Blocks(theme=gr.themes.Soft()) as demo:
+# Note: 'theme' removed from here per Gradio 6 migration guide
+with gr.Blocks() as demo:
     gr.Markdown("# 🎨 QuickDraw → tldraw JSON")
     
-    # --- FIX IS HERE: type='tuples' supports your current history format ---
-    chatbot = gr.Chatbot(height=500, type="tuples")
+    # Chatbot using default "messages" format (no type argument needed)
+    chatbot = gr.Chatbot(height=500)
     
     with gr.Row():
         img_input = gr.Image(type="pil", label="Upload Sketch", scale=1)
@@ -73,11 +85,18 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
             submit_btn = gr.Button("Send", variant="primary")
 
     def chat_wrapper(message, image, history):
+        # 1. Get response
         bot_res = respond(message, image, history)
-        history.append((message, bot_res))
+        
+        # 2. Update history using DICTIONARIES
+        history.append({"role": "user", "content": message})
+        history.append({"role": "assistant", "content": bot_res})
+        
         return "", None, history
 
+    # Initialize state as an empty list
     submit_btn.click(chat_wrapper, [txt_input, img_input, chatbot], [txt_input, img_input, chatbot])
     txt_input.submit(chat_wrapper, [txt_input, img_input, chatbot], [txt_input, img_input, chatbot])
 
-demo.launch()
+# Theme is now applied here in launch()
+demo.launch(theme=gr.themes.Soft())