Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Docker build fails on Apple Silicon #16

Open
eu9ene opened this issue Dec 22, 2021 · 6 comments
Open

Docker build fails on Apple Silicon #16

eu9ene opened this issue Dec 22, 2021 · 6 comments

Comments

@eu9ene
Copy link
Collaborator

eu9ene commented Dec 22, 2021

https://gist.github.com/mathjazz/46ca76c0136e7fd17ba9c3aba48951d6

@eu9ene
Copy link
Collaborator Author

eu9ene commented Dec 22, 2021

it is MBP with Apple Silicon, can we skip using intel MKL for Marian? https://marian-nmt.github.io/quickstart/

@eu9ene eu9ene closed this as completed Dec 22, 2021
@eu9ene eu9ene reopened this Dec 22, 2021
@andrenatal
Copy link
Contributor

I believe the option would be using openblas but @abhi-agg should know exactly what's required.

@eu9ene eu9ene changed the title Docker build fails Docker build fails on Apple Silicon Dec 23, 2021
@eu9ene
Copy link
Collaborator Author

eu9ene commented Dec 23, 2021

We can try but

Intel MKL is strongly recommended as it is faster

https://marian-nmt.github.io/docs/#cpu-version

@andrenatal
Copy link
Contributor

Is there another option?

@abhi-agg
Copy link
Collaborator

abhi-agg commented Jan 5, 2022

We can use OpenBlas instead of Intel MKL. It supports both intel and arm achitecture (see here).

We can do a bit more exploration to see how this can fit in docker and can work for both intel and arm architecture with the single docker image. But this is uncharted territory as we haven't used openblas ever and we don't know whether it works out of box in marian (we may see few surprises here).

But we should understand the use case first to start any effort on this.

@eu9ene
Copy link
Collaborator Author

eu9ene commented Jan 5, 2022

I think it makes sense to migrate to OpenBLAS taking into account Intel MKL licensing like @abhi-agg pointed out. We can benchmark performance using the load tests and see whether it is significantly slower.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

3 participants