Update app.py

app.py

@@ -43,6 +43,7 @@ model = AutoModelForCausalLM.from_pretrained(
     MODEL,
     torch_dtype=torch.bfloat16,
     device_map="auto",
+    trust_remote_code=True,
     quantization_config=quantization_config)
 
 @spaces.GPU(duration=100)