Skip to content

[PyTorch] Avoid registering FP8 scale update in ops without backward pass #2063

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
timmoon10 merged 2 commits into from
Aug 14, 2025
Merged

[PyTorch] Avoid registering FP8 scale update in ops without backward pass #2063

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
6a486e4
Avoid registering FP8 recipe update in ops without backward pass
timmoon10 Aug 12, 2025
fbccd45
Merge branch 'main' into debug-ops-with-no-bprop
timmoon10 Aug 12, 2025
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
7 changes: 6 additions & 1 deletion transformer_engine/pytorch/ops/fuser.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -176,14 +176,19 @@ def forward(
func_ctx.save_for_backward(*tensors_to_save)
func_ctx.tensor_objects = tensor_objects

# Whether to perform recipe update in backward pass
is_first_module = False
if fuser.first_op_requiring_backward < fuser._num_basic_ops:
is_first_module = FP8GlobalStateManager.is_first_fp8_module()

# Other context
func_ctx.backward_ops = fuser._backward_ops
func_ctx.basic_ops = fuser._basic_ops
func_ctx.basic_op_ctxs = basic_op_ctxs
func_ctx.basic_op_num_params = fuser._basic_op_num_params
func_ctx.num_extra_inputs = fuser.num_extra_inputs
func_ctx.num_extra_outputs = len(extra_outputs_flat)
func_ctx.is_first_module = FP8GlobalStateManager.is_first_fp8_module()
func_ctx.is_first_module = is_first_module

# Mark output tensors as not deletable in backward
for tensor in [x] + extra_outputs_flat:
Expand Down