Patrick Kidger comments

Results 1456 comments of


                                            Patrick Kidger

TypeCheckError: argument "<arg name>" (jaxlib.xla_extension.ArrayImpl) is not an instance of jax._src.prng.PRNGKeyArray

For anyone else coming across this: use `jaxtyping.PRNGKeyArray` and you should be good-to-go.

Upsampling/Unpooling for Transpose Convolutions

I think you're the first one to ask for them! :) I'd be happy to take a pull request that adds these.

Upsampling/Unpooling for Transpose Convolutions

Do you know how this operation is defined (mathemtically) in other frameworks e.g. PyTorch?

Upsampling/Unpooling for Transpose Convolutions

If this is given by a transpose, could this be implemented using `jax.linear_transpose`, without requiring any changes in JAX?

Upsampling/Unpooling for Transpose Convolutions

No opinions as I've not really used it I'm afraid :) Why would transposing be less efficient? XLA actually has a custom op (https://openxla.org/xla/operation_semantics#selectandscatter) for specifically the operation you're trying...

Possible regression with dataclasses validation

Thanks for the question! This is actually the reason that the breaking version number bumped with the v0.3.0 relese. This should now be described without the `size.` prefix: ```python @jaxtyped(typechecker=beartype)...

[Docos] Promote `ipython-beartype` with a trio of new FAQ entries on type-checking IPython REPLs, Jupyter Notebooks, and Google Colabs 🥳

Hehe, for completeness I should in turn acknowledge @knyazer, who actually put our IPython extension together! I just sat back and approved PRs 😁

[Docos] Promote `ipython-beartype` with a trio of new FAQ entries on type-checking IPython REPLs, Jupyter Notebooks, and Google Colabs 🥳

Ah yeah. FWIW I believe the impact of this bug is that debuggers (ipdb) will not work correctly whilst a magic is loaded.

[Docos] Promote `ipython-beartype` with a trio of new FAQ entries on type-checking IPython REPLs, Jupyter Notebooks, and Google Colabs 🥳

> jaxtyping shapes/dtypes too Does it work with checking consistency across multiple jaxtyping shapes/dtypes? (I.e. the bit that the `@jaxtyped` decorator is responsible for enforcing?)

Can shape information be preserved by the LSP ?

So I think this is just a choice that has been made in the implementation of the LSP -- since you're using VSCode I'm assuming it's probably pylance -- to...