mirror of
https://git.datalinker.icu/vllm-project/vllm.git
synced 2026-05-10 14:49:18 +08:00
77 lines
2.5 KiB
Python
77 lines
2.5 KiB
Python
from abc import ABC, abstractmethod
|
|
from typing import Dict, List, Optional
|
|
|
|
from vllm.config import (CacheConfig, DeviceConfig, LoRAConfig, ModelConfig,
|
|
ParallelConfig, SchedulerConfig, VisionLanguageConfig)
|
|
from vllm.lora.request import LoRARequest
|
|
from vllm.sequence import SamplerOutput, SequenceGroupMetadata
|
|
|
|
|
|
class ExecutorBase(ABC):
|
|
"""Base class for all executors.
|
|
|
|
An executor is responsible for executing the model on a specific device
|
|
type (e.g., CPU, GPU, Neuron, etc.). Or it can be a distributed executor
|
|
that can execute the model on multiple devices.
|
|
"""
|
|
|
|
@abstractmethod
|
|
def __init__(
|
|
self,
|
|
model_config: ModelConfig,
|
|
cache_config: CacheConfig,
|
|
parallel_config: ParallelConfig,
|
|
scheduler_config: SchedulerConfig,
|
|
device_config: DeviceConfig,
|
|
lora_config: Optional[LoRAConfig],
|
|
vision_language_config: Optional[VisionLanguageConfig],
|
|
) -> None:
|
|
raise NotImplementedError
|
|
|
|
@abstractmethod
|
|
def execute_model(self,
|
|
seq_group_metadata_list: List[SequenceGroupMetadata],
|
|
blocks_to_swap_in: Dict[int, int],
|
|
blocks_to_swap_out: Dict[int, int],
|
|
blocks_to_copy: Dict[int, List[int]]) -> SamplerOutput:
|
|
"""Executes one model step on the given sequences."""
|
|
raise NotImplementedError
|
|
|
|
@abstractmethod
|
|
def add_lora(self, lora_request: LoRARequest) -> bool:
|
|
raise NotImplementedError
|
|
|
|
@abstractmethod
|
|
def remove_lora(self, lora_id: int) -> bool:
|
|
raise NotImplementedError
|
|
|
|
@abstractmethod
|
|
def list_loras(self) -> List[int]:
|
|
raise NotImplementedError
|
|
|
|
@abstractmethod
|
|
def check_health(self) -> None:
|
|
"""Checks if the executor is healthy. If not, it should raise an
|
|
exception."""
|
|
raise NotImplementedError
|
|
|
|
|
|
class ExecutorAsyncBase(ExecutorBase):
|
|
|
|
@abstractmethod
|
|
async def execute_model_async(
|
|
self,
|
|
seq_group_metadata_list: List[SequenceGroupMetadata],
|
|
blocks_to_swap_in: Dict[int, int],
|
|
blocks_to_swap_out: Dict[int, int],
|
|
blocks_to_copy: Dict[int, List[int]],
|
|
) -> SamplerOutput:
|
|
"""Executes one model step on the given sequences."""
|
|
raise NotImplementedError
|
|
|
|
@abstractmethod
|
|
async def check_health_async(self) -> None:
|
|
"""Checks if the executor is healthy. If not, it should raise an
|
|
exception."""
|
|
raise NotImplementedError
|