Revert " Fp8 block scale quantization for fmha fwd (#3330)" (#3633)

This reverts commit dd0b4294af.
This commit is contained in:
Po Yen Chen
2026-01-23 13:21:19 +08:00
committed by GitHub
parent f30d04654e
commit de5a1d730d
14 changed files with 84 additions and 667 deletions

View File

@@ -12,7 +12,6 @@ enum class BlockAttentionQuantScaleEnum
{
NO_SCALE = 0,
PERTENSOR = 1,
BLOCKSCALE,
};
template <BlockAttentionQuantScaleEnum>
@@ -28,10 +27,5 @@ struct BlockAttentionQuantScaleEnumToStr<BlockAttentionQuantScaleEnum::PERTENSOR
{
static constexpr const char* name = "pertensor";
};
template <>
struct BlockAttentionQuantScaleEnumToStr<BlockAttentionQuantScaleEnum::BLOCKSCALE>
{
static constexpr const char* name = "blockscale";
};
} // namespace ck_tile