I tried to port a LSTM example in Python and the Rust version is much faster than the original. I wonder whether this is normal. 🤦