Run with M1 MacOS - Githubissues

abertsch72 / unlimiformer

Public repo for the NeurIPS 2023 paper "Unlimiformer: Long-Range Transformers with Unlimited Length Input"

MIT License

1.05k stars 80 forks source link

Run with M1 MacOS #6

Open TrieuLe0801 opened 1 year ago

TrieuLe0801 commented 1 year ago

Hello, How can it run on MacOS and M1 instead of GPU

urialon commented 1 year ago

Hi @TrieuLe0801 ,

Running on GPU or CPU is done in the exact same way! Did you try?

TrieuLe0801 commented 1 year ago

Hi @urialon I tried it on GPU already. I am trying to execute with M1 on MacOS, have unlimiformer supported for M1 yet?

urialon commented 1 year ago

Hi @TrieuLe0801 , There is no specific support for M1 - because it should be supported by the torch library. Everything is supposed to work the same way. Please let us know if you experience any problem.

Best, Uri

neubig commented 1 year ago

I think that this is referring to using the M1 GPU for accelerated training. I think there are two ways to do this in huggingface:

Use huggingface accelerate: https://huggingface.co/docs/accelerate/usage_guides/mps
Use PyTorch's .to("mps") call: https://huggingface.co/docs/diffusers/optimization/mps

I guess probably neither is supported in Unlimiformer at the moment (but it'd be great to have a community contribution for this!)