Fix MXFP4 expert zero weight loading

python/sglang/srt/layers/moe/fused_moe_triton/layer.py

@@ -727,7 +727,7 @@ def weight_loader(
         # if expert_id is None, then
         # all the experts are loaded at the same time
         if (
-            not expert_id
+            expert_id is None
             and self.quant_config is not None
             and self.quant_config.get_name() == "mxfp4"
             and self.quant_config.is_static_cfg()

test/registered/unit/layers/moe/test_mxfp4_expert_weight_loader.py

@@ -0,0 +1,52 @@
+from types import SimpleNamespace
+from unittest.mock import Mock, patch
+
+import torch
+
+from sglang.srt.layers.moe.fused_moe_triton.layer import FusedMoE
+from sglang.test.ci.ci_register import register_cpu_ci
+
+register_cpu_ci(est_time=3, suite="base-a-test-cpu")
+
+
+class _StaticMxfp4Config:
+    @staticmethod
+    def get_name():
+        return "mxfp4"
+
+    @staticmethod
+    def is_static_cfg():
+        return True
+
+
+def test_static_mxfp4_expert_zero_uses_per_expert_loader():
+    load_impl = Mock()
+    layer = SimpleNamespace(
+        quant_config=_StaticMxfp4Config(),
+        _map_global_expert_id_to_local_expert_id=lambda expert_id: expert_id,
+        _weight_loader_impl=load_impl,
+    )
+    param = torch.nn.Parameter(torch.empty(1))
+    loaded_weight = torch.empty((4, 8))
+
+    with patch(
+        "sglang.srt.layers.moe.fused_moe_triton.layer."
+        "get_global_expert_location_metadata",
+        return_value=None,
+    ):
+        FusedMoE.weight_loader(
+            layer,
+            param,
+            loaded_weight,
+            "experts.w1.weight",
+            "w1",
+            expert_id=0,
+        )
+
+    load_impl.assert_called_once_with(
+        param=param,
+        loaded_weight=loaded_weight,
+        weight_name="experts.w1.weight",
+        shard_id="w1",
+        expert_id=0,
+    )