Ray libraries

All Ray libraries run on a ray-vast cluster. Some need an extra change on vast. Each library page tells what works, what is different on vast, and which settings to use.

Send your work with ray job submit. Read Submit jobs.

Library	One worker	Many workers	What to know
Ray Core	✅	✅
Ray Data	🟨	🟨	Write the output to a cloud bucket. A local output path stays on each worker.
Ray Train	🟨	❌	Write checkpoints to a cloud bucket. Training across workers fails behind the NAT of vast. Use one worker with many GPUs.
Ray Tune	🟨	🟨	Write checkpoints to a cloud bucket.
Ray Serve	✅	✅
RLlib	✅	🟨	Put all learners on one worker. Learners across workers fail behind the NAT of vast.

✅ works. 🟨 works with an extra change. ❌ the main use fails.

Processes at once

Each worker runs a limited number of Ray processes (tasks and actors) at the same time. We call this number N. With the defaults of each library, your work fits. You need the settings of a library page only in special cases, for example many small tasks. Read Many tasks or actors.