diff --git a/tests/v1/spec_decode/test_tree_attention.py b/tests/v1/spec_decode/test_tree_attention.py index 42468daa62a9..456ce712d36e 100644 --- a/tests/v1/spec_decode/test_tree_attention.py +++ b/tests/v1/spec_decode/test_tree_attention.py @@ -155,7 +155,7 @@ def test_tree_attn_correctness() -> None: dim_per_head = 128 num_kv_heads = 2 - block_size = 128 + block_size = 32 max_sequence_length = 8192 randomize_blocks = True for batch_size in [1, 16, 32]: