biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[V1] Refactor get_executor_cls (#11754)

Browse Source
This commit is contained in:
Rui Qiao
2025-01-05 23:59:16 -08:00
committed by GitHub
parent f8fcca100b
commit 022c5c6944
5 changed files with 26 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
vllm/v1/engine/llm_engine.py
View File

@@ -89,7 +89,7 @@ class LLMEngine:
# Create the engine configs.
vllm_config = engine_args.create_engine_config(usage_context)
executor_class = cls._get_executor_cls(vllm_config)
executor_class = Executor.get_class(vllm_config)
if VLLM_ENABLE_V1_MULTIPROCESSING:
logger.debug("Enabling multiprocessing for LLMEngine.")
@@ -103,24 +103,6 @@ class LLMEngine:
stat_loggers=stat_loggers,
multiprocess_mode=enable_multiprocessing)
@classmethod
def _get_executor_cls(cls, vllm_config: VllmConfig) -> Type[Executor]:
executor_class: Type[Executor]
distributed_executor_backend = (
vllm_config.parallel_config.distributed_executor_backend)
if distributed_executor_backend == "ray":
from vllm.v1.executor.ray_executor import RayExecutor
executor_class = RayExecutor
elif distributed_executor_backend == "mp":
from vllm.v1.executor.multiproc_executor import MultiprocExecutor
executor_class = MultiprocExecutor
else:
assert (distributed_executor_backend is None)
from vllm.v1.executor.uniproc_executor import UniprocExecutor
executor_class = UniprocExecutor
return executor_class
def get_num_unfinished_requests(self) -> int:
return self.detokenizer.get_num_unfinished_requests()