Great work! Could you please provide inputs or insights on how we could run inference on ONNX? Thank you.