-
Hi devs,
I'm trying to `vmap` over an root finding procedure involving vector-valued parameters. I gather this is currently not possible with `jaxopt.ScipyRootFinding` because this is a wrapper aro…
-
I started looking at PR for https://github.com/patrick-kidger/equinox/issues/486. While doing so, I noticed another potential issue: [`is_array`'s `isinstance` check includes `np.generic`](https://gi…
-
I'm trying to differentiate the MJX step function via the autograd function `jax.grad()` in JAX, like:
```
def step(vel, pos):
mjx_data = mjx.make_data(mjx_model)
mjx_data = mjx_data.replace(q…
-
I used this installation environment without any errors throughout the process, and you can also do the same. I am using cuda11 and cudnn8.2, python version 3.8。jax_releases can found here https://sto…
-
Hey,
I can see how brax can be used to differentiate with respect to the system state. I wonder if there is a nice way to also diff wrt, e.g., the mass of a body. Taking the basic tutorial as an ex…
-
I often still write python as though it were FORTRAN, but there's no compiler to save me anymore. Hence, there are a bunch of places where relatively low-level python optimsation is possible.
For …
-
There seems to be a race to find the best implementation (currently 4) of KANs on the PyTorch side of the ML universe but there is a dearth of efforts on the JAX side (only 1 with flax). The flax impl…
-
### Description
`ReversibleSelect` seems to mess up JAX's backtracing/JIT compilation. In the code provided below, we define a simple (non reversible) model which (1) splits the inputs; (2) does so…
-
Hello everyone, practitioner here,
I am looking to train a very serious non-LLM model, and the training is expected to be very hard, so I am looking for maximum speed.
I know that Google's TPUs …
-
I was wondering is it possible to define both custom vjp and jvp for a function?
```
from jax import custom_vjp, custom_jvp, jacfwd, jacrev
def f(x):
return x
f = custom_jvp(f)
f.d…