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

Revert "stage3: efficient compute of scaled_global_grad_norm (#5256)" #5461

Merged
merged 1 commit into from
Apr 25, 2024
Merged

Revert "stage3: efficient compute of scaled_global_grad_norm (#5256)" #5461

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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: 2 additions & 2 deletions deepspeed/runtime/zero/stage3.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -15,7 +15,7 @@
from deepspeed.utils import logger
from deepspeed.runtime.fp16.loss_scaler import CreateLossScaler
from deepspeed.runtime.comm.coalesced_collectives import reduce_scatter_coalesced, all_to_all_quant_reduce
from deepspeed.runtime.utils import inf, is_model_parallel_parameter, get_only_unique_item
from deepspeed.runtime.utils import inf, get_global_norm, is_model_parallel_parameter, get_only_unique_item
from deepspeed.runtime.zero.partition_parameters import *
from deepspeed.runtime.zero.config import ZeroStageEnum
from deepspeed.runtime.zero.offload_config import OffloadDeviceEnum
Expand Down Expand Up @@ -2027,7 +2027,7 @@ def step(self, closure=None):
return

norm_groups = self._get_norm_groups()
scaled_global_grad_norm = torch.norm(torch.stack(norm_groups))
scaled_global_grad_norm = get_global_norm(norm_list=norm_groups)

# Stash unscaled gradient norm
self._global_grad_norm = scaled_global_grad_norm / self.loss_scale
Expand Down
Loading