Restruct sgl-kernel benchmark (#10861)

2025-09-25 07:45:25 +08:00
parent 7a06ef984d
commit c4e314f986
27 changed files with 425 additions and 319 deletions
--- a/python/sglang/srt/layers/quantization/w4afp8.py
+++ b/python/sglang/srt/layers/quantization/w4afp8.py
@@ -19,10 +19,6 @@ from sglang.srt.layers.quantization.unquant import UnquantizedLinearMethod
 from sglang.srt.layers.quantization.utils import is_layer_skipped
 from sglang.srt.utils import is_npu, set_weight_attrs

-_is_npu = is_npu()
-if not _is_npu:
-    from sglang.srt.layers.moe.cutlass_w4a8_moe import cutlass_w4a8_moe
-
 if TYPE_CHECKING:
    from sglang.srt.layers.moe import MoeRunnerConfig
    from sglang.srt.layers.moe.ep_moe.layer import EPMoE