Fix ZeRO stage 1 and add stage 2 support with DeepCompile #7366
loadams
merged
20 commits into
master
from
tohtana/dc_z1_no_sync
keep real inputs for partial recompilation
5921295c
Merge branch 'master' into tohtana/keep_real_inputs_for_recompile
ab9aad51
fix format
a7897cf2
keep gradient through gradient accumulation period
653cfded
rename functions to consolidate z1 and z2
6d9fedbe
add zero2
333a385f
add common functions
cdb54fdb
Merge branch 'master' of github.com:deepspeedai/DeepSpeed
0a2a71af
Merge branch 'master' into tohtana/dc_z0_no_sync
06237f05
Merge branch 'master' into tohtana/dc_z1_no_sync
021f3565
Merge branch 'master' into tohtana/dc_z1_no_sync
2a12ceac
Merge branch 'master' into tohtana/dc_z1_no_sync
1c44b8e5
fix release of ipg buffer
ae744bdd
remove name of unused variable
d77ae129
fix format
2d013237
Merge branch 'tohtana/fix_zero_bucket' into tohtana/dc_z1_no_sync
1829ec71
Merge branch 'master' into tohtana/dc_z1_no_sync
b6c70f92
Merge branch 'tohtana/dc_z1_no_sync' of github.com:deepspeedai/DeepSp…
33455cfb
Merge branch 'master' into tohtana/dc_z1_no_sync
1a94fbf8
Merge branch 'master' into tohtana/dc_z1_no_sync
44620054
loadams
approved these changes
on 2025-06-27
Login to write a write a comment.
Login via GitHub