Skip to content

Commit eb25f07

Browse files
authored
tests: xfail attention sink UT for sliding window + non causal case (#1752)
1 parent 9573c71 commit eb25f07

File tree

1 file changed

+5
-0
lines changed

1 file changed

+5
-0
lines changed

tests/test_attention_sink.py

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -635,6 +635,11 @@ def test_attention_sink_chunk_prefill(
635635
Simulate chunk-based processing of long sequences where current chunk
636636
attends to all historical tokens plus current chunk tokens
637637
"""
638+
if not causal and window_left >= 0:
639+
# xfail for non-causal + sliding window case
640+
pytest.xfail(
641+
"NOTE(Zihao): attention sink with sliding window and non-causal will fail after https://github.com/flashinfer-ai/flashinfer/pull/1661, temporarily xfail the test."
642+
)
638643
torch.manual_seed(42)
639644
device = torch.device("cuda:0")
640645
if backend == "fa3" and not is_sm90a_supported(device):

0 commit comments

Comments
 (0)