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 10, 07:45 UTC
[CI] fix `test_concat_and_cache_mla_rope_fused` (#32117)
Signed-off-by: zjy0516 <riverclouds.zhu@qq.com>
Jiangyun Zhu committed
Jan 11, 2026 at 23:11 UTC
3df619ac94bc60d14625cbc3405b051bc92c1c6e
1 file changed
+3
-2
tests/kernels/core/test_rotary_embedding_mla_cache_fused.py
+3
-2
@@ -13,7 +13,7 @@ from tests.kernels.allclose_default import get_default_atol, get_default_rtol
13
from tests.kernels.utils import DEFAULT_OPCHECK_TEST_UTILS, opcheck
14
from vllm import _custom_ops as ops
15
from vllm.model_executor.layers.rotary_embedding import RotaryEmbedding
16
-from vllm.platforms import current_platform
16
+from vllm.utils.torch_utils import set_random_seed
17
18
19
@pytest.mark.parametrize("dtype", [torch.half, torch.bfloat16, torch.float])
@@ -31,6 +31,7 @@ from vllm.platforms import current_platform
31
)
32
@torch.inference_mode()
33
def test_concat_and_cache_mla_rope_fused(
34
+ default_vllm_config,
35
dtype: torch.dtype,
36
is_neox_style: bool,
37
seq_len: int,
@@ -45,7 +46,7 @@ def test_concat_and_cache_mla_rope_fused(
46
max_position: int = 8192,
47
base: float = 10000,
48
) -> None:
48
- current_platform.seed_everything(seed)
49
+ set_random_seed(seed)
50
torch.set_default_device(device)
51
52
rope = RotaryEmbedding(