DeepSpeed
Keep fp32 masters on ZeRO-1 and ZeRO-2 when a resume skips optimizer state
#8891
Merged

Keep fp32 masters on ZeRO-1 and ZeRO-2 when a resume skips optimizer state #8891

sfc-gh-truwase
sfc-gh-truwase sfc-gh-truwase requested a review from tjruwase tjruwase 9 days ago
sfc-gh-truwase sfc-gh-truwase requested a review from tohtana tohtana 9 days ago
sfc-gh-truwase sfc-gh-truwase requested a review from loadams loadams 9 days ago
chatgpt-codex-connector
chatgpt-codex-connector
chatgpt-codex-connector commented on 2026-10-22
sfc-gh-truwase Keep fp32 masters on ZeRO-1 and ZeRO-2 when a resume skips optimizer …
f9958d4d
sfc-gh-truwase sfc-gh-truwase force pushed from 945aa1c8 to f9958d4d 9 days ago
sfc-gh-truwase sfc-gh-truwase requested a review from jeffra jeffra 9 days ago
sfc-gh-truwase sfc-gh-truwase changed the title Keep fp32 masters when a resume skips optimizer state Keep fp32 masters on ZeRO-1 and ZeRO-2 when a resume skips optimizer state 9 days ago
sfc-gh-truwase Merge master into the ZeRO-1/2 fp32 master branch.
b3728af3
sfc-gh-truwase sfc-gh-truwase requested a review from stas00 stas00 9 days ago
stas00
stas00 commented on 2026-10-22
sfc-gh-truwase Clarify FP32 checkpoint loading paths
e489f80b
sfc-gh-truwase Keep saved fp32 masters on every full ZeRO load.
a7ce02d9
sfc-gh-truwase Fail a universal load that is missing its fp32 master.
2f56cafc
sfc-gh-truwase Skip DistributedFixture tests when the world size is unavailable.
894a2b72
stas00
stas00 approved these changes on 2026-10-26
sfc-gh-truwase Merge branch 'master' into load-fp32-masters-without-optimizer-state
f02e7675
sfc-gh-truwase sfc-gh-truwase merged 2dd22dc1 into master 4 days ago
sfc-gh-truwase sfc-gh-truwase deleted the load-fp32-masters-without-optimizer-state branch 4 days ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone