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

Add Cache to Comm Group #4849

Open
wants to merge 6 commits into
base: master
Choose a base branch
from
Open

Add Cache to Comm Group #4849

Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
5e5cc84
Add cache to comm group
cmikeh2 Dec 20, 2023
9d176cf
Merge branch 'master' into cholmes/comm-group-cache
jomayeri May 23, 2024
89e7fb4
Merge branch 'master' into cholmes/comm-group-cache
jomayeri Sep 6, 2024
c3bbb12
Merge branch 'master' into cholmes/comm-group-cache
jomayeri Sep 9, 2024
efea0a8
Merge branch 'master' into cholmes/comm-group-cache
loadams Oct 28, 2024
cc9a57a
Merge branch 'master' into cholmes/comm-group-cache
loadams Jan 21, 2025
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
13 changes: 12 additions & 1 deletion deepspeed/comm/comm.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -52,6 +52,9 @@
ccl_backend = None
hccl_backend = None

# Cache of previously created comms
comms_cache = {}

# This should be set here so all rank/size information from the launcher can be propagated
from deepspeed.comm.utils import *

Expand Down Expand Up @@ -182,7 +185,15 @@ def new_group(ranks):
global cdb
assert cdb is not None and cdb.is_initialized(
), 'DeepSpeed backend not set, please initialize it using init_process_group()'
return cdb.new_group(ranks)

key = (cdb, tuple(ranks))

if key in comms_cache:
return comms_cache[key]
else:
new_group = cdb.new_group(ranks)
comms_cache[key] = new_group
return new_group


def is_available() -> bool:
Expand Down
Loading