DeepSpeed
Keep Yuan's o_proj bias whole on every rank and describe it as replicated
#8857
Merged
Overview
Commits
3
Changes
View On GitHub
Commits
  • Keep Yuan's o_proj bias whole on every rank and describe it as replicated
    Achyuthan-S committed 16 days ago
  • Run Yuan's o_proj bias through a real tensor-parallel group
    Achyuthan-S committed 16 days ago
  • Merge branch 'master' into step6-yuan-oproj-bias
    delock committed 14 days ago
FAQ Terms Privacy Refunds Impressum
Loading