Archis Joglekar

Thanks for the Q! The `concat` approach should work just fine but I agree that it's not necessarily the most transparent. Using a `pytree` based approach is likely most flexible...

Update here, `fixed_odeint` supports the state as a `dict` for now. We have yet to extend it to the adaptive solver.

Probably need a way to suppress it or provide a flag for usage because the NN usecase is such that this would be unnecessary at best and cumbersome at worst

A use case might be a long time-scale PDE solve, ~10^5 steps, 0.2-0.5s per step.

Hey, did you happen to make progress on this? I am curious to know and can hopefully provide some benchmarks when I get my problem running as well

for future readers who spend a few minutes trying to figure out what IFT is... IFT = Implicit Function Theorem

Here are some existing pytorch implementations for the enterprising scientific computing developer to follow https://github.com/xitorch/xitorch/blob/master/xitorch/interpolate/interp1.py https://github.com/patrick-kidger/torchcde/tree/master/torchcde

Thanks for your question! There was a new flag added to the fixed step solver (`fixed_odeint`) to indicate at which steps you would like to save the solution. The error...

I just took a look at the notebook and saw this error which is different than the assertion error you were getting earlier, so you probably have made some changes...

To echo @clemisch , many many thanks for that answer @mattjj . Really helpful to have that kind of insight !