biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Build/CI] Fix CUDA 11.8 build (#17679)

Browse Source 
Signed-off-by: Tyler Michael Smith <tyler@neuralmagic.com>
Signed-off-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
Signed-off-by: Tyler Michael Smith <tysmith@redhat.com>
Co-authored-by: Lucas Wilkinson <lwilkinson@neuralmagic.com>
This commit is contained in:
Tyler Michael Smith
2025-05-22 15:13:54 -04:00
committed by GitHub
parent f8d2cc5f55
commit 6e588da0f4
9 changed files with 78 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/kernels/moe/test_moe_permute_unpermute.py
View File

@@ -13,7 +13,7 @@ import torch
from vllm.model_executor.layers.fused_moe.fused_moe import fused_topk
from vllm.model_executor.layers.fused_moe.layer import determine_expert_map
from vllm.model_executor.layers.fused_moe.moe_permute_unpermute import (
moe_permute, moe_unpermute)
moe_permute, moe_permute_unpermute_supported, moe_unpermute)
from vllm.platforms import current_platform
NUM_EXPERTS = [16, 64]
@@ -167,6 +167,8 @@ def torch_unpermute(permuted_hidden_states: torch.Tensor,
def test_moe_permute_unpermute(n_token: int, n_hidden: int, topk: int,
n_expert: int, ep_size: int, dtype: torch.dtype,
align_block_size: Optional[int]):
if not moe_permute_unpermute_supported():
pytest.skip("moe_permute_unpermute is not supported on this platform.")
fill_invalid_expert = 0
ep_rank = np.random.randint(0, ep_size)
expert_map = None