fix

app.py

@@ -66,7 +66,7 @@ model = AutoModel.from_pretrained("{model_name}")"""
     return model_card
 
 def load_model(model_name, quantization_config, auth_token) : 
-    return AutoModel.from_pretrained(model_name, torch_dtype=torch.bfloat16, quantization_config=quantization_config, device_map="auto", use_auth_token=auth_token.token)
+    return AutoModel.from_pretrained(model_name, torch_dtype=torch.bfloat16, quantization_config=quantization_config, device_map="cpu", use_auth_token=auth_token.token)
 
 def load_model_cpu(model_name, quantization_config, auth_token) : 
     return AutoModel.from_pretrained(model_name, torch_dtype=torch.bfloat16, quantization_config=quantization_config, use_auth_token=auth_token.token)