Awni Hannun
Awni Hannun
I think wrapping cuFFT is a good solution for this https://developer.nvidia.com/cufft
Awesome, thanks!
Thanks for the PR! However, most of this functionality should already be included in https://github.com/ml-explore/mlx-examples/pull/1015 and https://github.com/ml-explore/mlx-examples/pull/1026, so I will close this. If there is anything here that those don't...
Happy to support you all here. We already have a T5 in MLX so that could serve as a good starting point https://github.com/ml-explore/mlx-examples/tree/main/t5 We would also gladly point people to...
That's awesome!!
In general we don't build or test MLX on anything less than Metal 3. Those older devices do not support metal 3 (see [table](https://developer.apple.com/metal/Metal-Feature-Set-Tables.pdf)) and I don't recommend using MLX...
Are you abele to provide some code that reproduces this? So far I'm not able to do so even when hitting that same kernel. This could also be something that's...
This one is actually not clear that it's fixed / not sure what the issue is here yet. The one that I think @angeloskath fixed in https://github.com/ml-explore/mlx/pull/2246 is this: ```...
We're not planning to change to C++23 anytime soon. We do intend to bump to C++20 soon though (maybe in the next week or so). Is it the case that...
I don't think that is accurate actually.. otherwise we'd have to compile every shared library we use with the same C++ version (which is definitely not the case).