Awni Hannun
Awni Hannun
Also useful if you can share MLX version and platform (OS, machine, etc)
Thanks for the code that's great. I'm running it now. It's at epoch 60 so far.. no segfault yet, let's see
> As you can see, the training speed in pytorch (about 8 s per epoch) is quite faster than that in mlx (about 12 s per epoch). Indeed I'm not...
I ran it for hundreds of epochs on both my M1 Max and an M2 Ultra and was not able to get a segfault. It may have been something we...
@JKwon0331 we aren't able to reproduce the segfault. Could you share a bit more information: 1. Operating system version 2. Output of `python -c "import mlx.core as mx; print(mx.__version__)"`
There are quite a few calls of `collapse_contiguous_dims` where the input strides do not have the same size. It looks like the function expects them to have the same size...
No not yet. Can you give rough estimates for how often and when the segfault shows up?
We believe this is fixed in the latest MLX (0.16). There was a bug in the part of the code that it seems you all were getting segfaults. Since we...
I use the HF CLI to login https://huggingface.co/docs/huggingface_hub/en/guides/cli#huggingface-cli-login Does that work for you?
> However, testing with progressively longer prompts reveals it begins producing nonsensical output (e.g., "...") after 8192 tokens, aligning with the "max_position_embeddings" value in the config.json file 🤔 not sure...