[Quantization] add humming mxfp4 moe backend (#41083)

Signed-off-by: Jinzhen Lin <jinzhen.ljz@antgroup.com>
This commit is contained in:
Jinzhen Lin
2026-05-02 18:36:03 -07:00
committed by GitHub
parent 856ec4804a
commit 08834cc3ce
9 changed files with 383 additions and 156 deletions
+2
View File
@@ -118,6 +118,7 @@ MoEBackend = Literal[
"flashinfer_cutlass",
"flashinfer_cutedsl",
"marlin",
"humming",
"aiter",
"emulation",
]
@@ -148,6 +149,7 @@ class KernelConfig:
- "flashinfer_cutlass": Use FlashInfer with CUTLASS kernels
- "flashinfer_cutedsl": Use FlashInfer with CuteDSL kernels (FP4 only)
- "marlin": Use Marlin kernels (weight-only quantization)
- "humming": Use Humming Mixed Precision kernels
- "aiter": Use AMD AITer kernels (ROCm only)
- "emulation": use BF16/FP16 GEMM, dequantizing weights and
running QDQ on activations.