chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

Revert "Support CuteDSL mm_fp4 backend" (#21077)

Browse Source
This commit is contained in:
Baizhou Zhang
2026-03-20 22:47:47 -07:00
committed by GitHub
parent c076968c52
commit 67cad3e69e
5 changed files with 219 additions and 109 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
python/sglang/srt/layers/quantization/fp4_utils.py
View File

@@ -18,7 +18,6 @@ class Fp4GemmRunnerBackend(Enum):
AUTO = "auto"
FLASHINFER_CUDNN = "flashinfer_cudnn"
FLASHINFER_CUTEDSL = "flashinfer_cutedsl"
FLASHINFER_CUTLASS = "flashinfer_cutlass"
FLASHINFER_TRTLLM = "flashinfer_trtllm"
@@ -34,9 +33,6 @@ class Fp4GemmRunnerBackend(Enum):
def is_flashinfer_trtllm(self) -> bool:
return self == Fp4GemmRunnerBackend.FLASHINFER_TRTLLM
def is_flashinfer_cutedsl(self) -> bool:
return self == Fp4GemmRunnerBackend.FLASHINFER_CUTEDSL
def get_flashinfer_backend(self) -> str:
"""Get the backend string to pass to FlashInfer's mm_fp4 API.
@@ -45,10 +41,7 @@ class Fp4GemmRunnerBackend(Enum):
'flashinfer_trtllm' -> 'trtllm'
'flashinfer_cutlass' -> 'cutlass'
'flashinfer_cudnn' -> 'cudnn'
'flashinfer_cutedsl' -> 'cute-dsl'
"""
if self == Fp4GemmRunnerBackend.FLASHINFER_CUTEDSL:
return "cute-dsl"
if self.value.startswith("flashinfer_"):
return self.value.removeprefix("flashinfer_")
else: