Update app.py

app.py

@@ -4,11 +4,10 @@ from peft import PeftModel
 import torch
 
 # --- CONFIGURATION ---
-BASE_MODEL = "unsloth/Qwen3-VL-2B-Instruct-unsloth-bnb-4bit" 
-LORA_ID = "EthanCastro/qwen3-vl-2b-quickdraw"
+BASE_MODEL = "unsloth/Qwen3-VL-2B-Instruct-unsloth-bnb-4bit"
+LORA_ID = "EthanCastro/qwen3-vl-2b-quickdraw"  # <--- Checked: Matches your repo name
 
 print("Loading model and processor...")
-# Load the base 4-bit model
 model = AutoModelForImageTextToText.from_pretrained(
     BASE_MODEL,
     torch_dtype=torch.float16,
@@ -16,7 +15,7 @@ model = AutoModelForImageTextToText.from_pretrained(
     trust_remote_code=True
 )
 
-# Apply your LoRA adapters from your Model Repo
+# 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!")
@@ -48,7 +47,7 @@ def respond(message, image, history):
     
     generated_text = processor.batch_decode(outputs, skip_special_tokens=True)[0]
     
-    # Extract only the latest assistant response
+    # Extract only the assistant's response
     if "assistant" in generated_text:
         response = generated_text.split("assistant")[-1].strip()
     else:
@@ -60,7 +59,8 @@ def respond(message, image, history):
 with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown("# 🎨 QuickDraw → tldraw JSON")
     
-    chatbot = gr.Chatbot(height=500)
+    # --- FIX IS HERE: type='tuples' supports your current history format ---
+    chatbot = gr.Chatbot(height=500, type="tuples")
     
     with gr.Row():
         img_input = gr.Image(type="pil", label="Upload Sketch", scale=1)