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

Upgrade Marian to the latest version #992

Open
eu9ene opened this issue Jan 15, 2025 · 3 comments
Open

Upgrade Marian to the latest version #992

eu9ene opened this issue Jan 15, 2025 · 3 comments

Comments

@eu9ene
Copy link
Collaborator

eu9ene commented Jan 15, 2025

No description provided.

@marco-c
Copy link
Collaborator

marco-c commented Jan 15, 2025

@ZJaume what were the recent improvements that could be good to use?

@ZJaume
Copy link
Collaborator

ZJaume commented Jan 15, 2025

I think:

  • Stable fp16 training.
  • "Compute 8.6 support if using CUDA>=11.1" I'm surprised using marian without this fix hasn't given any compilation issues on the cluster.
  • Ouptut sampling support, we could use it for backtranslation.
  • Parallel data loading.
  • Set different FFN size for the decoder.
  • Force decoding?
  • Some batch size related options to help training with more than 8 gpus.

See: https://github.com/marian-nmt/marian-dev/blob/master/CHANGELOG.md

@ZJaume
Copy link
Collaborator

ZJaume commented Jan 16, 2025

So it seems that I was wrong and the pipeline is using Marian v1.12. Fetched and installed from here:

repo: https://github.com/marian-nmt/marian-dev

and not using the submodule that is pointing to v1.10: https://github.com/mozilla/translations/tree/2eb36983958147fec3ab223827aea2084379bc15/3rd_party

So I guess this can be closed?

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