Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Jump to bottom

do allgather only in shared optimizer states groups #4167

Merged
merged 4 commits into from
Aug 21, 2023
Merged

do allgather only in shared optimizer states groups #4167

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
cd7d6b2
skip all-gather
inkcherry Aug 17, 2023
d2e0741
add notes
inkcherry Aug 17, 2023
e8985b6
Merge branch 'master' into allgather
loadams Aug 17, 2023
076a816
Merge branch 'master' into allgather
tjruwase Aug 21, 2023
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions deepspeed/runtime/utils.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -944,6 +944,10 @@ def all_gather_dp_groups(partitioned_param_groups, dp_process_group, start_align
partition_id = dist.get_rank(group=dp_process_group[group_id])
dp_world_size = dist.get_world_size(group=dp_process_group[group_id])

if dp_world_size == 1:
# no groups share optimizer states
# pipeline parallel with bf16 will default call this even if dp size = 1.
continue
num_shards = max(1, partitioned_params[partition_id].numel() * dp_world_size // allgather_bucket_size)

shard_size = partitioned_params[partition_id].numel() // num_shards
Expand Down
Loading