A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix] Fix test_mixtral_moe (#24371)
Jee Jee Li committed
Sep 7, 2025 at 00:32 UTC
7555d6b34af1aad14786a9451ff69968316a3ab4
1 file changed
+2
-2
tests/kernels/moe/test_moe.py
+2
-2
@@ -371,8 +371,8 @@ def test_fused_moe_wn16(m: int, n: int, k: int, e: int, topk: int,
371
@pytest.mark.parametrize(
372
"use_rocm_aiter", [True, False] if current_platform.is_rocm() else [False])
373
@torch.inference_mode()
374
-def test_mixtral_moe(dtype: torch.dtype, padding: bool, use_rocm_aiter: bool,
375
- monkeypatch):
374
+def test_mixtral_moe(dist_init, dtype: torch.dtype, padding: bool,
375
+ use_rocm_aiter: bool, monkeypatch):
376
"""Make sure our Mixtral MoE implementation agrees with the one from
377
huggingface."""
378