[QUESTION]Can the MoE layer be recomputed at a finer granularity inrecompute-modules
Open
community-request
module: moe
question
- Dominant language
- Python
- Stars
- 17.9k
- Forks
- 4.5k
- Avg merge
- 4d 6h
- Merged PRs (30d)
- 271
Description
When recomputing the MoE layer here, is it possible to support finer granularity — for example, by allowing recomputation of only linear_fc1 or linear_fc2?
Contributor guide
Assessment
This issue has not been assessed yet.