8000 cublaslt autotuning support for TunableOp by bilal2vec · Pull Request #133896 · pytorch/pytorch · GitHub

cublaslt autotuning support for TunableOp#133896

Closed

bilal2vec wants to merge 21 commits intopytorch:mainfrom

bilal2vec:bilal/cublas_tuning

Commits on Sep 4, 2024

cublaslt autotuning support for TunableOp

bilal2vec
authored and
pytorchmergebot
committed
add scaledgemm support

bilal2vec
authored and
pytorchmergebot
committed
enable autotuned scaledgemm for cuda

bilal2vec
authored and
pytorchmergebot
committed
add support for more fp8 dtypes (req for pytorch/cublaslt support) and fix fp8 dDesc issue

bilal2vec
authored and
pytorchmergebot
committed
address comments and lint

bilal2vec
authored and
pytorchmergebot
committed
add tunable gemm_and_bias

bilal2vec
authored and
pytorchmergebot
committed
replace elifdef with elif defined to pass rocm tests

bilal2vec
authored and
pytorchmergebot
committed
fix gemm_add_and_bias

bilal2vec
authored and
pytorchmergebot
committed
edit

bilal2vec
authored and
pytorchmergebot
committed
dispatch to cuda/rocm-specific fp8 types

bilal2vec
authored and
pytorchmergebot
committed
fix dispatching

bilal2vec
authored and
pytorchmergebot
committed
fix descriptors for bmm and scaled gemm

bilal2vec
authored and
pytorchmergebot
committed
add missing changes from last commit

bilal2vec
authored and
pytorchmergebot
committed
fix dispatching to fp8 dtypes for cuda/rocm

bilal2vec
authored and
pytorchmergebot
committed
lint

bilal2vec
authored and
pytorchmergebot
committed

Commits on Sep 21, 2024

lint
bilal2vec
committed

0