Awni Hannun

Results 1014 comments of Awni Hannun

Your description sounds conceptually correct to me. I'm curious how did you measure the memory use in your program (e.g. `# Memory: 1.88 GB`)? MLX has a memory buffer cache...

> Hmmm, do you know what triggers garbage collection (or where I could look in the repo to find out more)? [Here](https://github.com/ml-explore/mlx/blob/main/mlx/backend/metal/allocator.cpp#L187-L193) > Thinking about it some more, I suppose...

> So this might be somewhere on the to do list, albeit pretty far down, if it is? I would not say it's far down the list. This has been...

I think this is a bug in max and min 🤔 but I'm not sure if we should fix it. It will be tricky to fix in the general case...

@alxndrTL your comment is a different problem from this issue. You are taking and `exp(210)` which is `inf` which causes a `NaN` when you use `where`. I would not use...

Cool. I guess we can close this! Thanks for noticing it.

This is not a huge priority for us since it's a decent amount of work for something which is not super useful. We can update the docs for now until...

Yea it would be good to add the bitwise ops. `number_of_elements` probably shouldn't be added as it is mostly used internally for now.

David is working on one and we will add it to mlx-swift-examples when it's ready: https://github.com/ml-explore/mlx-swift-examples

@davidkoski still not ready for review?