diff --git a/tests/v1/e2e/general/test_mamba_prefix_cache.py b/tests/v1/e2e/general/test_mamba_prefix_cache.py index ceae041c6f..e857b12728 100644 --- a/tests/v1/e2e/general/test_mamba_prefix_cache.py +++ b/tests/v1/e2e/general/test_mamba_prefix_cache.py @@ -181,6 +181,7 @@ def get_fake_allocate_slots_fn(original_allocate_slots_fn: Callable): num_encoder_tokens: int = 0, full_sequence_must_fit: bool = False, reserved_blocks: int = 0, + has_scheduled_reqs: bool = True, ): ret = original_allocate_slots_fn( self, @@ -194,6 +195,7 @@ def get_fake_allocate_slots_fn(original_allocate_slots_fn: Callable): num_encoder_tokens, full_sequence_must_fit, reserved_blocks, + has_scheduled_reqs, ) if cur_step_action is not None: cur_block_ids = self.coordinator.single_type_managers[0].req_to_blocks[