Skip to content

[PyTorch] Use dummy amax for Float8Tensor cast #693

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
ksivaman merged 4 commits into from
Mar 1, 2024
Merged

[PyTorch] Use dummy amax for Float8Tensor cast #693

Changes from all commits
Commits
Show all changes
4 commits
Select commit Hold shift + click to select a range
4e88c88
Avoid updating real during param cast
ksivaman Feb 29, 2024
cb3e9a5
Merge branch 'main' into float8_tensor_amax_fix
ksivaman Feb 29, 2024
3306613
Review comments
ksivaman Mar 1, 2024
92434c1
Merge branch 'main' into float8_tensor_amax_fix
ksivaman Mar 1, 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
3 changes: 2 additions & 1 deletion transformer_engine/pytorch/module/base.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -790,7 +790,8 @@ def reset_parameters(self, defer_init: Optional[bool] = False) -> None:
param = Float8Tensor.to_float8(
param,
fp8_meta=self.fp8_meta,
fp8_meta_index=fp8_meta_index
fp8_meta_index=fp8_meta_index,
amax=torch.empty(1, device="cuda"), # Dummy amax to avoid overwriting history.
)

# Redo parameter wrap in case we broke it above
Expand Down