PR #10085 CUDA BFloat16 Refactor - SemanticDiff

Commits

Use BFLoat16 instead of nv_bfloat16 for CUDA

Lafi7e committed 4 years ago
fix BFloat16 rocm

mindest committed 4 years ago
remove gemm ut from rocm

Lafi7e committed 4 years ago
remove reducesum ut from rocm

Lafi7e committed 4 years ago
add biasgelu

Lafi7e committed 4 years ago
add support for more ops

Lafi7e committed 4 years ago
Merge branch 'master' into weicwang/bfloat16

Lafi7e committed 4 years ago
fix build error

Lafi7e committed 4 years ago
op md update

Lafi7e committed 4 years ago
Merge branch 'master' into weicwang/bfloat16

Lafi7e committed 4 years ago

Loading