Skip to content

Use torch._grouped_mm in eager mode #2721

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

Open
wujingyue wants to merge 2 commits into
base: main
Choose a base branch
from
Open

Use torch._grouped_mm in eager mode #2721

Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
ff9c62e
Use torch._grouped_mm in eager mode
wujingyue Nov 6, 2025
cbd61d9
Update layers_for_inference_benchmark.py
wujingyue Nov 11, 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
10 changes: 4 additions & 6 deletions thunder/benchmarks/layers_for_inference_benchmark.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -348,16 +348,14 @@ def _group_sizes_from_offsets(offsets: torch.Tensor) -> list[int]:


if LooseVersion(torch.__version__) >= LooseVersion("2.8.0"):
# Required otherwise, there is a graph-break.
# Required -- otherwise there is a graph-break.
_grouped_mm = torch.compiler.allow_in_graph(torch._grouped_mm)
else:
_grouped_mm = None


# This function should be replaced with torch._grouped_mm. However,
# torch._grouped_mm is yet to be usable because it requires offsets being
# multiples of 16.
def grouped_mm(a: torch.Tensor, b: torch.Tensor, offsets: torch.Tensor) -> torch.Tensor:
if torch.compiler.is_compiling():
# NOTE: This path also works for `thunder.jit` as it has a lookaside for `torch.compiler.is_compiling`.
if _grouped_mm is not None:
return _grouped_mm(a, b, offsets)

group_sizes = _group_sizes_from_offsets(offsets)
Expand Down
Loading