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

[CI Perf] Prune tests in tests/kernels/moe/ (#22939)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2025-08-14 23:33:42 -04:00
committed by GitHub
parent 590bddbfc5
commit d2b0e97ea6
6 changed files with 46 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
tests/kernels/moe/test_moe_permute_unpermute.py
View File

@@ -18,7 +18,7 @@ from vllm.model_executor.layers.fused_moe.moe_permute_unpermute import (
from vllm.platforms import current_platform
NUM_EXPERTS = [16, 64, 256]
TOP_KS = [2, 4, 6, 8]
TOP_KS = [2, 6, 8]
EP_SIZE = [1, 4, 16]
current_platform.seed_everything(0)
@@ -177,11 +177,11 @@ def torch_unpermute(permuted_hidden_states: torch.Tensor,
return output
@pytest.mark.parametrize("n_token", [1, 33, 64, 222, 1024, 2048, 3000, 5000])
@pytest.mark.parametrize("n_hidden", [2048, 4096, 7168])
@pytest.mark.parametrize("n_token", [1, 33, 1024, 5000])
@pytest.mark.parametrize("n_hidden", [2048, 7168])
@pytest.mark.parametrize("n_expert", NUM_EXPERTS)
@pytest.mark.parametrize("topk", TOP_KS)
@pytest.mark.parametrize("dtype", [torch.float16, torch.bfloat16])
@pytest.mark.parametrize("dtype", [torch.bfloat16])
@pytest.mark.parametrize("ep_size", EP_SIZE)
@pytest.mark.parametrize("align_block_size", [None, 128])
def test_moe_permute_unpermute(n_token: int, n_hidden: int, topk: int,