add seed in CPU UTs to avoid flaky failure (#7333)
This commit is contained in:
@@ -13,6 +13,8 @@ from sglang.srt.layers.moe.topk import grouped_topk_gpu as native_grouped_topk
|
||||
from sglang.srt.models.llama4 import Llama4MoE
|
||||
from sglang.test.test_utils import CustomTestCase
|
||||
|
||||
torch.manual_seed(0)
|
||||
|
||||
|
||||
# This is used by the Deepseek-V2 model
|
||||
class TestGroupedTopK(CustomTestCase):
|
||||
|
||||
Reference in New Issue
Block a user