GEMM + Swiglu fused Grouped MLP for MXFP8#2769

Merged

vthumbe1503 merged 76 commits intoNVIDIA:mainfrom

ksivaman:fused_mxfp8_grouped_mlp

Apr 3, 2026

Commits on Mar 17, 2026

Commits on Mar 30, 2026

Commits on Mar 31, 2026

Commits on Apr 1, 2026

Test only checks for fusion if fused kernel is available
timmoon10
committed
fix the tolerance to be of bf16 for the cute gemm
vthumbe1503
committed
Update transformer_engine/pytorch/ops/fused/forward_grouped_mlp.py

vthumbe1503
and
timmoon10
authored
address further review comments
vthumbe1503
committed
Merge branch 'fused_mxfp8_grouped_mlp' of github.com:ksivaman/TransformerEngine-1 into my_fused_grouped_mlp
vthumbe1503
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
address more review comments
vthumbe1503
committed
Merge branch 'fused_mxfp8_grouped_mlp' of github.com:ksivaman/TransformerEngine-1 into my_fused_grouped_mlp
vthumbe1503
committed
address more review comments + test for zero grouped tensor work case
vthumbe1503
committed
cublaslt remove zero work gemm avoidance
vthumbe1503
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
address review comments
vthumbe1503
committed
Merge branch 'fused_mxfp8_grouped_mlp' of github.com:ksivaman/TransformerEngine-1 into my_fused_grouped_mlp
vthumbe1503
committed
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
fix the wgrad test
vthumbe1503
committed
Merge branch 'fused_mxfp8_grouped_mlp' of github.com:ksivaman/TransformerEngine-1 into my_fused_grouped_mlp
vthumbe1503
committed
split dbias functionality from gq api
ksivaman
committed

Commits on Apr 2, 2026