EngineX-Hygon/sglang
5
0
Fork 0
Code Issues Pull Requests Actions 7 Projects Releases Wiki Activity

Fix circular import (#10107)

Browse Source
This commit is contained in:
Cheng Wan
2025-09-06 01:34:17 -07:00
committed by GitHub
parent 21af5c0404
commit a5a03209e9
5 changed files with 25 additions and 33 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
python/sglang/srt/layers/quantization/w4afp8.py
View File

@@ -9,7 +9,6 @@ from torch.nn.parameter import Parameter
from sglang.srt.distributed.parallel_state import get_moe_expert_parallel_world_size
from sglang.srt.layers.linear import LinearBase, UnquantizedLinearMethod
from sglang.srt.layers.moe.cutlass_w4a8_moe import cutlass_w4a8_moe
from sglang.srt.layers.quantization.base_config import (
FusedMoEMethodBase,
QuantizationConfig,
@@ -297,6 +296,7 @@ class W4AFp8MoEMethod(FusedMoEMethodBase):
dispatch_output: StandardDispatchOutput,
) -> CombineInput:
from sglang.srt.layers.moe.cutlass_w4a8_moe import cutlass_w4a8_moe
from sglang.srt.layers.moe.token_dispatcher import StandardCombineInput
x = dispatch_output.hidden_states