DeepSpeed
ZeRO Gradient Accumulation Dtype.
#2847
Merged

ZeRO Gradient Accumulation Dtype. #2847

jomayeri merged 60 commits into master from jomayeri/new-zero-accum
jomayeri
jomayeri Adding attributes for grad accum dtype.
faa8e16c
jomayeri accumulating reduction grads in stage 2 mode 2
f7c35615
jomayeri jomayeri requested a review from jeffra jeffra 2 years ago
jomayeri jomayeri requested a review from tjruwase tjruwase 2 years ago
jomayeri jomayeri requested a review from samyam samyam 2 years ago
jomayeri jomayeri requested a review from mrwyattii mrwyattii 2 years ago
jomayeri missing colon
04cd6a74
jomayeri tracking reduc grad move
4b878848
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
8570eca7
tjruwase
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
b48629c5
stas00
jomayeri Correct hooks.
33f9d203
jomayeri Merge remote-tracking branch 'refs/remotes/origin/jomayeri/new-zero-a…
543a7d3a
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
9022ce37
stas00
stas00 commented on 2023-02-21
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
f137bfa4
jomayeri Name change updates.
31a16447
jomayeri Merge remote-tracking branch 'refs/remotes/origin/jomayeri/new-zero-a…
bb9b6a69
stas00
jomayeri Using grad_accum in cpu offload functions.
a7d6b60a
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
9c6c4d36
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
f6f36724
tjruwase
tjruwase commented on 2023-02-24
tjruwase
tjruwase commented on 2023-02-24
stas00
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
63e6cc4f
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
2ca8199c
jomayeri Addressing comments: putting bf opt back, removing hooks
ef80b4c4
jomayeri Fixing missing pointer to grad accum.
4e0f920b
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
a9d86e40
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
fa406b14
tjruwase
tjruwase commented on 2023-03-01
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
c3fff11b
tjruwase
tjruwase commented on 2023-03-01
jomayeri Renaming functions.
09b2eea9
danyang-rainbow
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
aabe5a5e
jomayeri Merge remote-tracking branch 'refs/remotes/origin/jomayeri/new-zero-a…
73860770
jomayeri More function renames.
d65907a0
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
96d37af2
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
7cf1ddbf
jomayeri Adding reduction dtype.
e0eb4e63
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
814c20a3
tjruwase
tjruwase commented on 2023-03-22
tjruwase
tjruwase commented on 2023-03-22
tjruwase
tjruwase commented on 2023-03-22
tjruwase
tjruwase commented on 2023-03-22
tjruwase
tjruwase commented on 2023-03-22
tjruwase
tjruwase commented on 2023-03-22
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
5c9d4f58
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
0e4271ad
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
e783894b
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
830180eb
jomayeri Merge branch 'jomayeri/new-zero-accum' of github.com:microsoft/DeepSp…
cd30ca2c
danyang-rainbow
tjruwase Merge branch 'master' into jomayeri/new-zero-accum
9c7074d7
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
cce6cf16
jomayeri updating for offload
d741446c
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
c5387fee
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
ac0a73c8
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
3a2280f5
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
8c3bfe1d
jomayeri Adding functionality for stage 3.
b2c5f5ce
jomayeri jomayeri changed the title ZeRO 1 and 2 Gradient Accumulation Dtype. ZeRO Gradient Accumulation Dtype. 2 years ago
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
25d42962
jomayeri Adding s3 test support.
49623cad
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
15ebfda1
jomayeri Add to MiCS optimizer.
1d31297c
HeyangQin zero++ tutorial PR (#3783)
df1859d6
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
22fa5591
jeffra jeffra force-pushed the master branch from f6e2e38b to bafaf3c0 2 years ago
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
6cd7a222
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
993ddc1e
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
df29a1e2
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
94ab3b87
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
6cad2a25
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
62a82368
jomayeri Removing need to grad_reduc attribute.
26cefc57
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
e2129f0a
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
88f6aa2e
tjruwase tjruwase requested a review from tjruwase tjruwase 2 years ago
tjruwase
tjruwase approved these changes on 2023-07-19
jomayeri Offload correctness.
0fab8df6
jomayeri Merge branch 'master' into jomayeri/new-zero-accum
8492da7d
jomayeri jomayeri enabled auto-merge 2 years ago
jomayeri jomayeri merged 8afcda2a into master 2 years ago
zaptrem
tjruwase
zaptrem

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone