Subchannel Block quantized GEMM - #1545

Merged

timmoon10 merged 23 commits into

NVIDIA:mainfrom

kwyss-nvidia:kwyss/cublas_gemm_github_mr

Apr 7, 2025

Commits on Apr 4, 2025

Add GEMM logic for blockwise quantized tensors.
kwyss-nvidia
committed
fbcbcb0
Update NVTE_BLOCK_SCALING for GEMM.
kwyss-nvidia
committed
522ffbe
Gate feature on CUDA 12.9
kwyss-nvidia
committed
d7e1fce
Gemm typo.
kwyss-nvidia
committed
f212c81
Remove unecessary type converter change.
kwyss-nvidia
committed
48b2d57
Reflect epilogue availability and test supported epilogues.
kwyss-nvidia
committed
5761589
GEMM simplifications from recipe branch.
kwyss-nvidia
committed
07b19b7
Format py code.
kwyss-nvidia
committed
c4a41b8
Update GEMM DGelu tests to match support depending on output dtype.
kwyss-nvidia
committed
51ed2fb
Force pow2Scales in GEMM
kwyss-nvidia
committed
e7af140
Add GEMM test to pytorch test suite.
kwyss-nvidia
committed
596a009
Add copyright to GEMM test.
kwyss-nvidia
committed
4aa6067
Update import for GEMM test.
kwyss-nvidia
committed
758dc4a
Add license.
kwyss-nvidia
committed
7d5b5d9
Update test gemm supported predicate.
kwyss-nvidia
committed
efdf8e0

Commits on Apr 5, 2025

Use sgemm like interfaces and naming.
kwyss-nvidia
committed
a9f209a
Rewrite GEMM comment.
kwyss-nvidia
committed
861c870
MR Feedback.
kwyss-nvidia
committed
ada6438

Commits on Apr 6, 2025

Refactor GEMM param canonicalization
timmoon10
committed
e484269
[pre-commit.ci] auto fixes from pre-commit.com hooks
pre-commit-ci[bot]
committed
9f0707e
Merge branch 'main' into kwyss/cublas_gemm_github_mr
timmoon10
authored
cf36b99

Commits on Apr 7, 2025

Prune number of tests.
kwyss-nvidia
committed
f3123cf
Merge branch 'main' into kwyss/cublas_gemm_github_mr
timmoon10
authored
4e4c59e