Skip to content

[PyTorch] Propagate FP8 graph weight update flag in GroupedLinear #3052

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 3 commits into from
Jun 5, 2026
+10 −1
Merged

[PyTorch] Propagate FP8 graph weight update flag in GroupedLinear #3052

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
e445000
Fix GroupedLinear FP8 graph weight update flag
allenphilipj Jun 4, 2026
fc56af4
Merge branch 'main' into codex-grouped-linear-fp8-cudagraph-skip
allenphilipj Jun 4, 2026
7e8020c
Merge branch 'main' into codex-grouped-linear-fp8-cudagraph-skip
ksivaman Jun 4, 2026
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
11 changes: 10 additions & 1 deletion transformer_engine/pytorch/module/grouped_linear.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -1696,6 +1696,15 @@ def forward(
f"does not match number of GEMMs ({num_gemms})."
)

if FP8GlobalStateManager.fp8_graph_capturing():
skip_fp8_weight_update = (
FP8GlobalStateManager.quantization_state.skip_fp8_weight_update_tensor
)
else:
skip_fp8_weight_update = None
if skip_fp8_weight_update is not None:
is_first_microbatch = False

# Preprocess input tensor
if isinstance(inp, QuantizedTensorStorage):
raise TypeError("GroupedLinear doesn't support input tensor in FP8.")
Expand Down Expand Up @@ -1754,7 +1763,7 @@ def forward(
is_grad_enabled,
weight_workspaces,
cache_weight,
None, # skip_fp8_weight_update
skip_fp8_weight_update,
self.save_original_input,
debug,
)
Expand Down
Loading