Fix CompressedTensorsMxInt4MoE abstract method and relax GPQA baseline (#19726)

Co-authored-by: Alison Shao <alisonshao@Mac.attlocal.net>
2026-03-02 19:03:21 -08:00
parent 8dfb6e1684
commit fe9d85d93c
2 changed files with 2 additions and 2 deletions
--- a/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_mxint4_moe.py
+++ b/python/sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w4a4_mxint4_moe.py
@@ -287,7 +287,7 @@ class CompressedTensorsMxInt4MoE(CompressedTensorsMoEScheme):
    ):
        self.moe_runner_config = moe_runner_config

-    def apply(
+    def apply_weights(
        self,
        layer: torch.nn.Module,
        dispatch_output: StandardDispatchOutput,