[V1] Why there is no swap queue in V1 version Scheduler?

[Ghjk94522]

As mentioned in the title, there is no swap deque in class Scheduler, and no swap relative operation in schedule func.
Is this not implemented yet or just remove for some reason?

class Scheduler:
    def __init__(
        self,
        scheduler_config: SchedulerConfig,
        cache_config: CacheConfig,
        lora_config: Optional[LoRAConfig],
    ) -> None:
        ...
        self.requests: Dict[str, Request] = {}
        # Priority queues for requests.
        self.waiting: Deque[Request] = deque()
        self.running: List[Request] = []
        ...

codes in the vllm/v1/core/scheduler.py shown as above

And I found the _initialize_kv_caches func in class EngineCore init the cpu blocks number to 0 as followed:

 def _initialize_kv_caches(self,
                              cache_config: CacheConfig) -> Tuple[int, int]:
        num_gpu_blocks, _ = self.model_executor.determine_num_available_blocks(
        )
        ...
        num_cpu_blocks = 0
        self.model_executor.initialize(num_gpu_blocks)
        return num_gpu_blocks, num_cpu_blocks

Maybe someone can help me to explain this?

[comaniac]

Swapping in v1 is removed intentionally. Instead, we expect when recomputing the preempted requests, most prompt tokens could be bypassed due to prefix caching.

[Ghjk94522]

get, thanks