[core][scheduler] simplify and improve scheduler (#6867)
This commit is contained in:
@@ -183,7 +183,7 @@ def test_v1_v2_greedy_equality_with_cow(baseline_llm_generator,
|
||||
|
||||
# Allow only 2 sequences of ~128 tokens in worst case.
|
||||
# Note 16 = 128/block_size
|
||||
"num_gpu_blocks_override": 2 * (16 + 1),
|
||||
"num_gpu_blocks_override": 2 * (16 + 2),
|
||||
}
|
||||
])
|
||||
@pytest.mark.parametrize("baseline_llm_kwargs", [{
|
||||
|
||||
Reference in New Issue
Block a user