Carlos Martin

Great. Once #898 is merged I'll put together a PR.

@mattjj Any update on this? If I understand correctly, `JAX_PLATFORM_NAME` is deprecated in favor of `JAX_PLATFORMS`. However, there's still [some code](https://github.com/search?q=repo%3Agoogle/jax%20jax_platform_name&type=code) that uses the former. This includes the code that...

@lanctot Is this wontfix? Just wondering.

@lanctot No problem. I wrote a self-contained module of JAX-compatible helper functions for OpenSpiel games.

I also just encountered this issue. @SteveDiamond @rileyjmurray Could we reopen https://github.com/cvxpy/cvxpy/pull/1174?

@SteveDiamond Thanks. Do you have an outline of the changes that are needed?

Will multidimensional variables ever be added to cvxpy?

There's a JAX implementation of [tfp.mcmc.sample_halton_sequence](https://www.tensorflow.org/probability/api_docs/python/tfp/mcmc/sample_halton_sequence): [tfp.substrates.jax.mcmc.sample_halton_sequence](https://www.tensorflow.org/probability/api_docs/python/tfp/substrates/jax/mcmc/sample_halton_sequence). Is there an analogous JAX implementation of [tf.math.sobol_sample](https://www.tensorflow.org/api_docs/python/tf/math/sobol_sample)? I don't see one under [tfp.substrates.jax.math](https://www.tensorflow.org/probability/api_docs/python/tfp/substrates/jax/math).

@jakevdp > For GPU backends, each iteration effectively requires a kernel launch Is this a limitation of XLA or a fundamental limitation of the hardware itself? (I also asked this...

@cgarciae Isn't shape inference from inputs, as done for the `inputs` argument, more in line with flax's `init` philosophy? The `RNN` situation could be resolved as follows: 1. Add a...