An experimental framework to train LRU networks in pytorch (linear recurrent unit networks, see Orvieto et al., Resurrecting Recurrent Neural Networks for Long Sequences, 2023), and fit dynamical systems with them, among other tasks.
Also supporting material for my blog post RNNs strike back in the notebook ds_exps.ipynb.
Code adapted from the jax/flax implementation of N. Zucchet minimal-LRU.
Still unstable and may change significantly.