[Bugfix] Fix test_mixtral_moe (vllm-project#24371)

Author: jeejeelee

tests/kernels/moe/test_moe.py

@@ -371,8 +371,8 @@ def test_fused_moe_wn16(m: int, n: int, k: int, e: int, topk: int,
 @pytest.mark.parametrize(
     "use_rocm_aiter", [True, False] if current_platform.is_rocm() else [False])
 @torch.inference_mode()
-def test_mixtral_moe(dtype: torch.dtype, padding: bool, use_rocm_aiter: bool,
-                     monkeypatch):
+def test_mixtral_moe(dist_init, dtype: torch.dtype, padding: bool,
+                     use_rocm_aiter: bool, monkeypatch):
     """Make sure our Mixtral MoE implementation agrees with the one from
     huggingface."""