acsweet

Not sure if you solved this yet (and I'm no expert in this!), but maybe you can try something like this? `python -m torch.distributed.launch --nproc_per_node 4 -m vall_e.train yaml=config/your_data/ar_or_nar.yml`

I'd like to pick up on this issue (first time contributor) starting with `fft` if that's okay

I'm going to start with the "easy" stuff already implemented in mlx, and I'll start in `mlx.math` with - `fft2` - `rfft` - `irfft` - `qr` (I'll have to see...

@awni Thank you! I'll keep you updated as I progress. Right now, would it be possible to get `stft` and `istft` implemented on the mlx side? It looks like it...

I'm going to hold off on `math.qr` for now, mlx currently only supports square matrices (and no option for the complete or reduced factorization). I have a PR for `fft2`,...

I'm going to start working through `mlx.nn` now. I hope that's okay, but I'm going to start with `conv`, and if @lkarthee or @Faisal-Alsrheed would like to jump back in,...

@awni Would it be possible to get support for non-square matrices implemented in `mlx.linalg.qr`? I didn't see an open issue for it, I can open a feature enhancement too.

If the `conv` implementation looks good, I think I'll get started on the other convolutional functions - `depthwise_conv` - `separable_conv` - `conv_transpose`

I have a pull request for the remaining convolutional functions, if those look good I'll continue! Fadi asked to work on `max_pool` and `avg_pool`, so I'm going to work on...

@awni Would it be pretty straightforward to implement singular value norms in `linalg::matrix_norm`? I can open an issue for it too!