Awni Hannun comments

Results 1014 comments of


                                            Awni Hannun

[BUG] c++ compile cache should be thread safe

> Inner state of mlx::core should be thread safe There are still a few places this is not true still. Off the top of my head: - rng state -...

Array API backends support for MLX

Indeed, MLX plans to support the array API with two exceptions: 1. float64 2. Ops with data-dependent output shapes I think you'll find those exceptions will not be uncommon (e.g....

Array API backends support for MLX

That would be great! And functions which you think MLX should have that are part of the array API but missing we can add. Just let us know.

Add utilities to distributions

We don't have a `distributions` package in the same way the PyTorch does and don't currently have a plan yet to add one to MLX core. We can keep this...

scatter operations missing index locations

There's no plan to add this at the moment. A `scatter_argmax` might be doable, but I'm curious what would you use it for?

libc++abi: terminating due to uncaught exception of type std::runtime_error: [Event::stream] Cannot access stream on invalid event.

Can you share code that reproduces the error you're seeing?

libc++abi: terminating due to uncaught exception of type std::runtime_error: [Event::stream] Cannot access stream on invalid event.

Unfortunately that snippet isn’t sufficient for us to reproduce and debug this. If you can provide a standalone program that we can run that reproduces the issue that would be...

libc++abi: terminating due to uncaught exception of type std::runtime_error: [Event::stream] Cannot access stream on invalid event.

Closing as inactive.

[BUG] `Invalid type ndarray` for complex, non-contiguous `np.array` input to `mx.array`

This looks like a Nanobind issue. FIled here https://github.com/wjakob/nanobind/issues/709, let's see if it is something they can fix.

Remove unused import

I'm not following exactly what the purpose of this diff is. If you import the subpackage members without putting them in `__all__` then you get a warning?