The human-centric platform for production ML & AI

Outerbounds

hey folks, is there a writeup around @parallel <https://github.com/Netflix/metaflow/issues/444#issuecomment-1489000064> decorator somewhere?

re: writeup - if you are interested in an example - here is <https://github.com/Netflix/metaflow/blob/master/test/parallel/parallel_test_flow.py|one>

We don't have anything written up yet (but a good time to start documenting it). We are happy to answer any specific questions. <@U04KGH31TTM> and his team has been using it for some time now - they may have more real world insights to share.

got it! I'll have a few use cases for it too.
Has anyone tried fine-tuning LLM using this new feature?

Hello <@U05ASJYD0KX>,

We have been using `@parallel` for a variety of different use cases. I've used it with PyTorch Lightning + Deepspeed to fine-tune an LLM, and <@U04LJA2JDEV> is currently experimenting with `@parallel` and FSDP and to fine-tune a T5-3b model. Would be happy to help with any questions you might have.