Awni Hannun comments

Results 1014 comments of


                                            Awni Hannun

mistral example seems to hang on Loading model from disk.

@Blucknote There are pre-converted quantized models in the MLX Hugging Face community: https://huggingface.co/mlx-community Also, all of the conversion scripts in the [LLM examples](https://github.com/ml-explore/mlx-examples/tree/main/llms) can produce quantized models

[BUG] cifar example fails with ValueError when replacing LayerNorm with BatchNorm with default params

That is not expected, it sounds like a bug. Thanks for reporting, I will take a look.

[BUG] cifar example fails with ValueError when replacing LayerNorm with BatchNorm with default params

Yes we are very much aware of this issue. Working with @angeloskath on a fix.

[BUG] cifar example fails with ValueError when replacing LayerNorm with BatchNorm with default params

For now I recommend avoiding the running stats until we fix it

[BUG] cifar example fails with ValueError when replacing LayerNorm with BatchNorm with default params

This should fix it https://github.com/ml-explore/mlx/pull/385

LoRA finetuning with model downloaded from mlx-community fails

Thanks @singhaki for the input. Regarding the rope_traditional flag, I wouldn't remove it as the model won't work as well. We can add the config param to our model so...

LoRA finetuning with model downloaded from mlx-community fails

This will be updated in #252

generate.py bug: using global var args.prompt instead of prompt passed to the method

Sorry about that, the fix is in #227

LoRA training really slow

It’s probably using too much memory. Read the [section in the readme](https://github.com/ml-explore/mlx-examples/tree/main/lora#memory-issues) on how to reduce memory use. If it’s still super slow your machine may not have enough memory...

LoRA training really slow

Wow that's odd. Did you happen to do anything on your computer around the time that it slowed down? It's possible the GPU got used by something else? Also you...