Skip to content

Use correct default for round robin gradients #1258

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

Merged
tjruwase merged 6 commits into from
Jul 28, 2021
Merged

Use correct default for round robin gradients #1258

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
b9fe6bb
Make round robin gradient partitioning configurable (default False)
tjruwase Jul 28, 2021
88dc448
Use the correct default
tjruwase Jul 28, 2021
273b45f
Merge with master
tjruwase Jul 28, 2021
8326d3b
Merge branch 'master' into olruwase/round_robin_gradient_option
tjruwase Jul 28, 2021
cf4b34f
Log config setting
tjruwase Jul 28, 2021
92d27e3
Merge branch 'olruwase/round_robin_gradient_option' of github.com:mic…
tjruwase Jul 28, 2021
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
2 changes: 1 addition & 1 deletion deepspeed/runtime/zero/config.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -189,4 +189,4 @@ def _initialize(self, zero_config_dict):
self.round_robin_gradients = get_scalar_param(
zero_config_dict,
ZERO_OPTIMIZATION_ROUND_ROBIN_GRADIENTS,
ZERO3_OPTIMIZATION_CONTIGUOUS_GRADIENTS_DEFAULT)
ZERO_OPTIMIZATION_ROUND_ROBIN_GRADIENTS_DEFAULT)
1 change: 1 addition & 0 deletions deepspeed/runtime/zero/stage2.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,7 @@ def __init__(self,
logger.info(f"Reduce bucket size {reduce_bucket_size}")
logger.info(f"Allgather bucket size {allgather_bucket_size}")
logger.info(f"CPU Offload: {cpu_offload}")
logger.info(f'Round robin gradient partitioning: {round_robin_gradients}')
# The fused optimizer does all the work. We need this layer for two reason:
# 1. maintain same user API from apex.fp16_utils
# 2. keep common stuff here in case we need to add ne552w fused optimizer later
Expand Down