Patrick Kidger

Do note that such a tag describes the structure of the particular root find problem that's being tackled, which is kind of complicated. That aside, it might be easier for...

Right, but - at least in principle - one could provide a solver that perform a tridiagonal solve whilst completely ignoring tags. I'm just concious that the tags here would...

Hey there! So the main option for controlling memory usage is the `RecursiveCheckpointAdjoint(checkpoints=...)`. Smaller values increase computation and decrease memory usage. If this is not provided explicitly (the default) then...

1. Reversible solves are currently not implemented in Diffrax (though we have a PR to change that, you note!) 2. If you don't backpropagate then any choice of adjoint will...

Hi there! This is something which can be done pretty straightforwardly with a custom solver. For example supposing your slow dynamics are independent of your fast dynamics, then (untested) it...

Hi there! So if you want a Stratonovich solution and implicit-drift + implicit diffusion, then you can probably just use the existing implicit ODE solvers. For the more common 'implicit...

I'm afraid it gets worse than that! For example, `jax.tree.map(lambda _: object(), some_module)` will replace all attributes with an `object()`. Or to use another common example, `jax.vmap(..., in_axes=..., out_axes=...)` require...

Correct!

I honestly don't have super strong feelings. `jax.extend` doesn't actually come with any greater stability guarantees than e.g. `jax._src`, after all! :)

Returning to just the 'field' case, it would usually be preferable to implement this as a `@property` that forwards on, just like the method case. I think the issue you're...