[question] Migrate from Cloud Run Jobs to Worker Pools

[supreme-gg-gg]

Preview feature, I think this can make the training, inference, evaluation / export more performant and efficient because they're all highly reliant on jobs right now. For training, this is a hard limitation since we cannot use multi GPU for distributed training.

I'm not entirely sure since this is a new feature but this sounds like a cool thing to try

https://cloud.google.com/run/docs/deploy-worker-pools