DeepSpeed
MoE dispatch: fix silent Tutel + TP corruption, speed up native and Tutel paths
#8195
Merged
Overview
Commits
6
Changes
View On GitHub
Commits
  • Speed up Tutel route extraction for k > 2
    jinyouzhi committed 17 days ago
  • Shard the Tutel dispatch buffer along capacity under tensor parallelism
    jinyouzhi committed 17 days ago
  • Dispatch MoE tokens by index instead of a dense one-hot einsum
    jinyouzhi committed 17 days ago
  • Merge branch 'master' into moe
    tohtana committed 16 days ago
  • Merge remote-tracking branch 'origin/master' into HEAD
    jinyouzhi committed 13 days ago
  • Move MoE unit tests to tests/unit/v1/moe
    jinyouzhi committed 13 days ago
FAQ Terms Privacy Refunds Impressum
Loading