Skip to content

Wire MoE and MXFP8Linear training paths to CuTeDSL by default #4620

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

Closed
alexsamardzic wants to merge 2 commits into from
Closed

Wire MoE and MXFP8Linear training paths to CuTeDSL by default #4620

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
0cd0647
Update
alexsamardzic Jul 27, 2026
864b33b
Update
alexsamardzic Jul 29, 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
8 changes: 8 additions & 0 deletions benchmarks/prototype/moe_training/mxfp8/bench_ep_pipeline.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -40,6 +40,7 @@
from torchao.prototype.moe_training.ep.unpermute import _unpermute_bf16
from torchao.prototype.moe_training.mxfp8_grouped_mm import (
_to_mxfp8_then_scaled_grouped_mm,
set_mxfp8_grouped_mm_backend,
)

device = torch.device("cuda")
Expand Down Expand Up @@ -554,6 +555,7 @@ def print_results(experiments: List[Experiment]):
def main(args: argparse.Namespace):
"""Main benchmark entry point."""
torch.random.manual_seed(123)
set_mxfp8_grouped_mm_backend(args.backend)

# Set up process group
setup_distributed()
Expand Down Expand Up @@ -597,5 +599,11 @@ def setup_distributed():
action="store_true",
help="Use torch.compile",
)
parser.add_argument(
"--backend",
choices=("legacy", "cutedsl"),
default="cutedsl",
help="MXFP8 grouped-mm backend",
)
args = parser.parse_args()
main(args)
Loading
Loading