Issue with trying the Phi-2 model about mlx-examples HOT 6 CLOSED

This is super odd, you shouldn't even need TensorFlow for this example, it's not a dependency.

Maybe try updating your Transformers:

pip install -U transformers

from mlx-examples.

I tried to update the transformers, but I still encountered the same error.

from mlx-examples.

This seems like your environment corrupted or in a wonky state. Are you able to do anything with mlx? Like what happens if you do:

import mlx
a = mlx.zeros((2, 2))
print(a)

Maybe try uninstalling TensorFlow and/or making a clean conda env?

from mlx-examples.

I have also issues with running the convert.py -

it requires CUDA for the flash_attn module, which is not available obviously on my m1 macbook pro.
is there a way to publish the .npz files with the weights on huggingface? or to workaround this issue?

thanks

from mlx-examples.

I think you are using an old version of Transformers try: pip install -U transformers.

Actually it has a minimal version on it. Make sure you do pip install -r requirements.txt before trying to run the example. I will close this, let me know if the issue persists and I can reopen.

from mlx-examples.

That fixed the problem for me, thank you very much for the fast help!

from mlx-examples.