DeepSpeed
ZeRO3, improved parameter all-gather operation
#1188
Merged

ZeRO3, improved parameter all-gather operation #1188

zarzen
zarzen remove norm(), avoid memcpy after allgather
1e73e758
zarzen zarzen requested a review from arashashari arashashari 4 years ago
zarzen zarzen requested a review from awan-10 awan-10 4 years ago
zarzen zarzen requested a review from cli99 cli99 4 years ago
zarzen zarzen requested a review from conglongli conglongli 4 years ago
zarzen zarzen requested a review from eltonzheng eltonzheng 4 years ago
zarzen zarzen requested a review from jeffra jeffra 4 years ago
zarzen zarzen requested a review from minjiaz minjiaz 4 years ago
zarzen zarzen requested a review from niumanar niumanar 4 years ago
zarzen zarzen requested a review from RezaYazdaniAminabadi RezaYazdaniAminabadi 4 years ago
zarzen zarzen requested a review from samyam samyam 4 years ago
zarzen zarzen requested a review from ShadenSmith ShadenSmith 4 years ago
zarzen zarzen requested a review from tjruwase tjruwase 4 years ago
ghost
zarzen zarzen marked this pull request as draft 4 years ago
zarzen zarzen force pushed to 1e73e758 4 years ago
zarzen WIP: wrapped ncclAllgather as customized op in DS
67b3db3e
zarzen WIP: integrated into partition_parameters
70e681f0
zarzen
zarzen Fix format
81b4fc4a
zarzen zarzen marked this pull request as ready for review 4 years ago
zarzen Merge branch 'master' into impr_allgather_params
8a14e434
zarzen cleaned dead code, modified unit test
32c8fa72
zarzen Merge branch 'master' into impr_allgather_params
c4728f50
tjruwase Merge branch 'master' into impr_allgather_params
e075fd4d
tjruwase tjruwase removed review request from conglongli conglongli 4 years ago
tjruwase tjruwase removed review request from awan-10 awan-10 4 years ago
tjruwase tjruwase removed review request from arashashari arashashari 4 years ago
tjruwase tjruwase removed review request from cli99 cli99 4 years ago
tjruwase tjruwase removed review request from eltonzheng eltonzheng 4 years ago
tjruwase tjruwase removed review request from minjiaz minjiaz 4 years ago
tjruwase tjruwase removed review request from RezaYazdaniAminabadi RezaYazdaniAminabadi 4 years ago
tjruwase tjruwase removed review request from niumanar niumanar 4 years ago
zarzen
tjruwase
tjruwase
zarzen
zarzen removed customized c++ extension
52085080
zarzen Merge remote-tracking branch 'origin/master' into impr_allgather_params
ffd3d3b0
zarzen change torch.ones to torch empty
1ed96ce2
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
220f2e0e
tjruwase Merge branch 'master' into impr_allgather_params
8f655940
zarzen typo
0e6d8e0c
tjruwase Merge branch 'master' into impr_allgather_params
691749fe
tjruwase Merge branch 'master' into impr_allgather_params
88e750e6
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
497ee7d7
tjruwase
zarzen
tjruwase
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
bd8839c5
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
25829106
tjruwase Merge branch 'master' into impr_allgather_params
4ca0d391
tjruwase Merge branch 'master' into impr_allgather_params
56de9ad9
tjruwase Merge branch 'master' into impr_allgather_params
056cf101
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
aac09cd2
tjruwase
jeffra
jeffra
jeffra commented on 2021-10-07
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
6201b295
tjruwase
zarzen
zarzen warn if not cuda tensor for allgather
50a9215d
zarzen
jeffra Merge branch 'master' into impr_allgather_params
c554a589
tjruwase Merge branch 'master' into impr_allgather_params
b7e131d4
zarzen fix formatting
813cb227
tjruwase Merge branch 'master' into impr_allgather_params
588d3d0e
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
eb0a540f
zarzen fix: move ds_tensor to cuda device
c092b789
zarzen
tjruwase
tjruwase Merge branch 'master' into impr_allgather_params
e73809dc
zarzen
tjruwase Merge branch 'master' into impr_allgather_params
d1d3c28f
tjruwase
tjruwase commented on 2021-10-27
tjruwase
tjruwase commented on 2021-10-27
tjruwase Merge branch 'master' into impr_allgather_params
62cb1049
tjruwase Merge branch 'master' into impr_allgather_params
ab64b17a
zarzen remove try clause on the path for fetching params
7a801721
zarzen Merge branch 'microsoft:master' into impr_allgather_params
f01dad85
tjruwase
tjruwase approved these changes on 2021-10-30
tjruwase Merge branch 'master' into impr_allgather_params
524e6096
tjruwase Merge branch 'master' into impr_allgather_params
d7fff585
tjruwase tjruwase enabled auto-merge (squash) 4 years ago
tjruwase tjruwase merged c0eeb69d into master 4 years ago

Login to write a write a comment.

Login via GitHub

Assignees
No one assigned
Labels
Milestone