819a4459ce
### What this PR does / why we need it?
Drop vLLM 0.13.0 support, upgrade to 0.14.0
- vLLM version: v0.13.0
- vLLM main:
d68209402d
---------
Signed-off-by: hfadzxy <starmoon_zhang@163.com>
68 lines
2.6 KiB
Python
68 lines
2.6 KiB
Python
from collections.abc import Iterator
|
|
from typing import Optional
|
|
|
|
import torch
|
|
from vllm.config import VllmConfig
|
|
from vllm.v1.attention.backend import AttentionBackend # type: ignore
|
|
from vllm.v1.kv_offload.abstract import LoadStoreSpec, OffloadingManager
|
|
from vllm.v1.kv_offload.backends.cpu import CPUBackend
|
|
from vllm.v1.kv_offload.lru_manager import LRUOffloadingManager
|
|
from vllm.v1.kv_offload.mediums import CPULoadStoreSpec, GPULoadStoreSpec
|
|
from vllm.v1.kv_offload.spec import OffloadingSpec
|
|
from vllm.v1.kv_offload.worker.worker import OffloadingHandler
|
|
from vllm.v1.kv_cache_interface import KVCacheConfig
|
|
|
|
from vllm_ascend.kv_offload.cpu_npu import CpuNpuOffloadingHandler
|
|
|
|
|
|
class NPUOffloadingSpec(OffloadingSpec):
|
|
|
|
def __init__(self,
|
|
vllm_config: VllmConfig,
|
|
kv_cache_config: Optional[KVCacheConfig] = None):
|
|
super().__init__(vllm_config, kv_cache_config)
|
|
|
|
num_cpu_blocks = self.extra_config.get("num_cpu_blocks")
|
|
if not num_cpu_blocks:
|
|
raise Exception(
|
|
"num_cpu_blocks must be specified in kv_connector_extra_config"
|
|
)
|
|
self.num_cpu_blocks: int = num_cpu_blocks
|
|
|
|
# scheduler-side
|
|
self._manager: Optional[OffloadingManager] = None
|
|
|
|
# worker-side
|
|
self._handler: Optional[OffloadingHandler] = None
|
|
|
|
def get_manager(self) -> OffloadingManager:
|
|
if not self._manager:
|
|
kv_events_config = self.vllm_config.kv_events_config
|
|
enable_events = (kv_events_config is not None
|
|
and kv_events_config.enable_kv_cache_events)
|
|
self._manager = LRUOffloadingManager(
|
|
CPUBackend(block_size=self.offloaded_block_size,
|
|
num_blocks=self.num_cpu_blocks),
|
|
enable_events=enable_events,
|
|
)
|
|
return self._manager
|
|
|
|
def get_handlers(
|
|
self,
|
|
kv_caches: dict[str, torch.Tensor],
|
|
attn_backends: dict[str, type[AttentionBackend]],
|
|
) -> Iterator[tuple[type[LoadStoreSpec], type[LoadStoreSpec],
|
|
OffloadingHandler]]:
|
|
if not self._handler:
|
|
self._handler = CpuNpuOffloadingHandler(
|
|
attn_backends=attn_backends,
|
|
gpu_block_size=self.gpu_block_size,
|
|
cpu_block_size=self.offloaded_block_size,
|
|
num_cpu_blocks=self.num_cpu_blocks,
|
|
gpu_caches=kv_caches,
|
|
)
|
|
|
|
assert self._handler is not None
|
|
yield GPULoadStoreSpec, CPULoadStoreSpec, self._handler
|
|
yield CPULoadStoreSpec, GPULoadStoreSpec, self._handler
|