[BugFix] Avoid prefix cache hit in the same schedule step for mamba layers (#29387)

Signed-off-by: Chen Zhang <zhangch99@outlook.com>
This commit is contained in:
Chen Zhang
2026-02-10 07:41:16 +00:00
committed by GitHub
parent dab1de9f38
commit 97fa8f6590
6 changed files with 78 additions and 3 deletions
+2
View File
@@ -857,6 +857,8 @@ def test_prefill_hybrid_model_combinations(spec_types: list[str]):
# Should have blocks for all groups
assert len(blocks.get_block_ids()) == num_groups
manager.new_step_starts()
# Second request: should hit cached blocks for common prefix
req1 = make_request("1", common_token_ids + [4] * 5, block_size, hash_fn)
computed_blocks, num_computed_tokens = manager.get_computed_blocks(req1)