[BugFix] Fix FA3 IMA with FULL_AND_PIECEWISE and cascade attention (default) (#28702)

2025-11-14 07:19:22 -05:00
parent 9324e10275
commit db56a59970
2 changed files with 5 additions and 2 deletions
--- a/tests/kernels/attention/test_cascade_flash_attn.py
+++ b/tests/kernels/attention/test_cascade_flash_attn.py
@@ -170,6 +170,7 @@ def test_cascade(
        logits_soft_cap=soft_cap if soft_cap is not None else 0,
        block_table=block_tables,
        common_prefix_len=common_prefix_len,
+        max_num_splits=0,  # no max
        fa_version=fa_version,
    )