deas-mhumhna

> Indeed, a code snippet to reproduce the issue would be great. https://colab.research.google.com/drive/1c9kkaAwrysPFq5bEPBhJrzQDC6sGT0sF?usp=sharing

Thanks for your response. I refactored the code for `chebval` and `chebder` to be more `jit` friendly; the original functions were direct copies from the current NumPy implementations with minor...

Thanks for the quick response. I was using `jax.experimental.ode.odeint` before I found Diffrax and I can't remember the exact difference between `train_step` and `eval_step`. I switched because `odeint` was taking...