DeepSpeed
Unmanaged gradient accumulation: ZeRO offload support
#8225
Merged

Unmanaged gradient accumulation: ZeRO offload support #8225

sfc-gh-truwase
sfc-gh-truwase Extend unmanaged gradient accumulation to ZeRO offload
d3dcf9ad
sfc-gh-truwase sfc-gh-truwase requested a review from tjruwase tjruwase 16 days ago
sfc-gh-truwase sfc-gh-truwase requested a review from tohtana tohtana 16 days ago
sfc-gh-truwase sfc-gh-truwase requested a review from loadams loadams 16 days ago
chatgpt-codex-connector
chatgpt-codex-connector commented on 2026-08-06
sfc-gh-truwase sfc-gh-truwase requested a review from stas00 stas00 16 days ago
sfc-gh-truwase sfc-gh-truwase requested a review from delock delock 16 days ago
sfc-gh-truwase Skip stale offload grads for inactive params in unmanaged finalize
e425f836
sfc-gh-truwase Add unmanaged ZeRO-3 NVMe offload equivalence tests
636d8f2a
sfc-gh-truwase Merge branch 'master' into sfc-gh-truwase/gas_mgmt_offload
925fe4c7
delock
delock commented on 2026-08-08
delock
delock commented on 2026-08-08
delock
delock
delock approved these changes on 2026-08-08
delock
delock commented on 2026-08-08
delock
delock requested changes on 2026-08-08
sfc-gh-truwase Use getter/setter for ZeRO GAS boundary flag
fbed95e4
sfc-gh-truwase
delock
delock approved these changes on 2026-08-09
delock Merge branch 'master' into sfc-gh-truwase/gas_mgmt_offload
6b6a5269
sfc-gh-truwase sfc-gh-truwase merged cf443004 into master 13 days ago
sfc-gh-truwase sfc-gh-truwase deleted the sfc-gh-truwase/gas_mgmt_offload branch 13 days ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone