Commit 4b535e64 authored by zhangshao's avatar zhangshao
Browse files

update

parent 34e67b1e
...@@ -1820,7 +1820,7 @@ def test_flash_attn_output_mla( ...@@ -1820,7 +1820,7 @@ def test_flash_attn_output_mla(
# @pytest.mark.parametrize("d", [32, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256]) # @pytest.mark.parametrize("d", [32, 59, 64, 80, 96, 111, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256]) # @pytest.mark.parametrize("d", [32, 64, 96, 128, 160, 192, 224, 256])
# @pytest.mark.parametrize("d", [160, 192, 224, 256]) # @pytest.mark.parametrize("d", [160, 192, 224, 256])
@pytest.mark.parametrize('d', [128]) @pytest.mark.parametrize('d', [64, 128])
@pytest.mark.parametrize( @pytest.mark.parametrize(
"seqlen_q,seqlen_k", "seqlen_q,seqlen_k",
[ [
......
...@@ -1261,7 +1261,7 @@ def cal_diff(x: torch.Tensor, y: torch.Tensor, name: str, use_fp8: bool = False, ...@@ -1261,7 +1261,7 @@ def cal_diff(x: torch.Tensor, y: torch.Tensor, name: str, use_fp8: bool = False,
@pytest.mark.parametrize("use_qq_bias", [True, False]) # seqlen_q > seqlen_k 时 skip @pytest.mark.parametrize("use_qq_bias", [True, False]) # seqlen_q > seqlen_k 时 skip
@pytest.mark.parametrize("use_sinks", [True, False]) @pytest.mark.parametrize("use_sinks", [True, False])
@pytest.mark.parametrize("use_mm_prefix", [True, False]) @pytest.mark.parametrize("use_mm_prefix", [True, False])
@pytest.mark.parametrize("d", [256]) @pytest.mark.parametrize("d", [128, 256])
@pytest.mark.parametrize( @pytest.mark.parametrize(
"batch_size,seqlen_q,seqlen_k,block_size", "batch_size,seqlen_q,seqlen_k,block_size",
[ [
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment