AI-Hypercomputer / AI-Hypercomputer/JetStream
MOE with JetStream
Open
- Dominant language
- Python
- Stars
- 457
- Forks
- 67
- PR merge metrics
- No merged PRs in 30d
Description
Could someone explain or point to a doc that explains how MOE is implemented on Jetstream? Specifically, the all-to-all communications, static vs dynamic, sparse matmuls.
I would like to understand how XLA compiles MOE.
Contributor guide
Assessment
This issue has not been assessed yet.