Vincent Roulet

Hello @hjia, Thanks for reporting this! The issue is that `lower` and `upper` need to be static arguments for making the implicit differentiation compatible with jit (in the `solve_2` and...

We could want to make the Bisection method differentiable with respect to its lower and upper values. The following code fails for example. But so it would be nice to...

Yes, I see the issue. This would be good to know. Thanks!

I'm now printing the type of the input: - If the function is jitted: - At compilation time, this type must be a Tracer. - Afterwards, nothing will be printed....

> the idea is that having a callable for a step-size could be used not just for simple pre-defined schedules, but also for line-searches like (say) a custom Armijo line-search...

Hello @carlosgmartin, Wouldn't [jaxprs](https://jax.readthedocs.io/en/latest/jaxpr.html) be what you are looking for? JAX traces the whole computational graph so that you could directly work on it. I may not have got what...

So no, we haven't that now and we were not planning on adding it. Wouldn't a package like https://github.com/cvxgrp/cvxpylayers be a good starting point?

Hello! Could you send a minimal code example? I'm happy to investigate that. Which method are you using?

Thanks! For the first error, it's just that to find a root of a function, this function must be from your input space to your input space. Right now you...

Hello Justin! Yes, the issue is that pure_callback would require the whole code inside to be in a numpy format. But since the function is supposed to be in jax,...