[Bugfix] Initialize DeepGemmQuantScaleFMT oracle lazily; bound QuantFP8 UE8M0 packed path to group_size 128 (#51359)

Signed-off-by: BabyDrangoner <148877251+BabyDrangoner@users.noreply.github.com>
Signed-off-by: kiroxu <148877251+BabyDrangoner@users.noreply.github.com>
Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: Wentao Ye <44945378+yewentao256@users.noreply.github.com>
Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
kiroxu
2026-08-12 14:27:50 -04:00
committed by GitHub
co-authored by Copilot Wentao Ye mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent d20a031d33
commit 19d61b1273
3 changed files with 30 additions and 2 deletions
@@ -94,6 +94,8 @@ class QuantFP8(CustomOp):
self.is_group_quant
and self.use_ue8m0
and self.use_deep_gemm_supported
and not self.static
and self.group_size == 128
and (DeepGemmQuantScaleFMT.from_oracle() == DeepGemmQuantScaleFMT.UE8M0)
):
return fp8_utils.per_token_group_quant_fp8_packed_for_deepgemm(