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

Switch to torch.compile for dropout for torch v>2.2 #607

Merged
merged 2 commits into from
Jan 19, 2024
Merged

Switch to torch.compile for dropout for torch v>2.2 #607

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
7c6b929
Switch to torch.compile for dropout for torch v>2.2
ksivaman Jan 16, 2024
a4d4450
Merge branch 'main' into fix_torch_compile_dropout_perf
ksivaman Jan 17, 2024
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: 7 additions & 2 deletions transformer_engine/pytorch/jit.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -12,6 +12,11 @@
if torch.__version__ >= "2" and bool(int(os.getenv("NVTE_TORCH_COMPILE", "1"))):
jit_fuser = torch.compile

# See: https://github.com/NVIDIA/TransformerEngine/issues/597
dropout_fuser = torch.jit.script
if torch.__version__ >= "2.2" and bool(int(os.getenv("NVTE_TORCH_COMPILE", "1"))):
dropout_fuser = torch.compile

# Decorator to disable Torch Dynamo
# See: https://github.com/NVIDIA/TransformerEngine/issues/308
no_torch_dynamo = lambda recursive=True: lambda func: func
Expand Down Expand Up @@ -134,7 +139,7 @@ def _bias_dropout_add(x, bias, residual, prob):
return _bias_dropout_add


@jit_fuser
@dropout_fuser
def bias_dropout_add_fused_train_(
x: torch.Tensor, bias: torch.Tensor, residual: torch.Tensor, prob: float
) -> torch.Tensor:
Expand All @@ -151,7 +156,7 @@ def bias_dropout_add_fused_train(
return bias_dropout_add_fused_train_(x, bias, residual, prob)


@jit_fuser
@dropout_fuser
def bias_dropout_add_fused_inference_(
x: torch.Tensor, bias: torch.Tensor, residual: torch.Tensor, prob: float
) -> torch.Tensor:
Expand Down