Skip to content

remove duplicate check for pp and zero stage #4033

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
mrwyattii merged 4 commits into from
Jul 25, 2023
Merged

remove duplicate check for pp and zero stage #4033

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
742083d
remove duplicate check for pp and zero stage
inkcherry Jul 25, 2023
1cb4ca8
remove line
inkcherry Jul 25, 2023
a48c8a6
use ZeroStageEnum
inkcherry Jul 25, 2023
4b7108a
Merge branch 'master' into remove_check
tjruwase Jul 25, 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
9 changes: 4 additions & 5 deletions deepspeed/runtime/pipe/engine.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -55,7 +55,8 @@ def __init__(self, has_bool_tensors=False, *super_args, **super_kwargs):
super().__init__(*super_args, **super_kwargs)
assert isinstance(self.module, PipelineModule), "model must base PipelineModule"

assert self.zero_optimization_stage() < 2, "ZeRO-2 and ZeRO-3 are incompatible with pipeline parallelism"
assert self.zero_optimization_stage(
) < ZeroStageEnum.gradients, "ZeRO-2 and ZeRO-3 are incompatible with pipeline parallelism"

# We schedule the all-reduces, so disable it in super().backward()
self.enable_backward_allreduce = False
Expand Down Expand Up @@ -243,10 +244,8 @@ def _exec_reduce_grads(self):
self._force_grad_boundary = True
if self.pipeline_enable_backward_allreduce:
if self.bfloat16_enabled():
if self.zero_optimization_stage() < ZeroStageEnum.gradients:
self._bf16_reduce_grads()
else:
raise NotImplementedError("PP+BF16 only work for ZeRO Stage 1")
# PP+BF16 work for ZeRO Stage 1
self._bf16_reduce_grads()
else:
self.allreduce_gradients(bucket_size=MEMORY_OPT_ALLREDUCE_SIZE)
self._force_grad_boundary = False
Expand Down