ONNX version of message-intent model. Will be used on GPU machine.