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

[V1] V1 Enablement Oracle (#13726)

Browse Source 
Signed-off-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Co-authored-by: rshaw@neuralmagic.com <rshaw@neuralmagic.com>
Co-authored-by: Nicolò Lucchesi <nlucches@redhat.com>
Co-authored-by: Tyler Michael Smith <tyler@neuralmagic.com>
Co-authored-by: Michael Goin <michael@neuralmagic.com>
This commit is contained in:
Robert Shaw
2025-03-15 01:02:20 -04:00
committed by GitHub
parent 8c0d15d5c5
commit d4d93db2c5
96 changed files with 1537 additions and 512 deletions
Download Patch File Download Diff File Expand all files Collapse all files

53
vllm/engine/async_llm_engine.py
View File

@@ -595,6 +595,13 @@ class AsyncLLMEngine(EngineClient):
log_requests: bool = True,
start_engine_loop: bool = True,
**kwargs) -> None:
if envs.VLLM_USE_V1:
raise ValueError(
"Using V0 AsyncLLMEngine, but envs.VLLM_USE_V1=True. "
"This should not happen. As a workaround, try using "
"AsyncLLMEngine.from_vllm_config(...) or explicitly set "
"VLLM_USE_V1=0 or 1 and report this issue on Github.")
self.log_requests = log_requests
self.engine = self._engine_class(*args, **kwargs)
@@ -629,33 +636,53 @@ class AsyncLLMEngine(EngineClient):
engine_config: VllmConfig) -> Type[ExecutorBase]:
return LLMEngine._get_executor_cls(engine_config)
@classmethod
def from_vllm_config(
cls,
vllm_config: VllmConfig,
start_engine_loop: bool = True,
usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
stat_loggers: Optional[dict[str, StatLoggerBase]] = None,
disable_log_requests: bool = False,
disable_log_stats: bool = False,
) -> "AsyncLLMEngine":
"""Create an AsyncLLMEngine from the EngineArgs."""
return cls(
vllm_config=vllm_config,
executor_class=cls._get_executor_cls(vllm_config),
start_engine_loop=start_engine_loop,
log_requests=not disable_log_requests,
log_stats=not disable_log_stats,
usage_context=usage_context,
stat_loggers=stat_loggers,
)
@classmethod
def from_engine_args(
cls,
engine_args: AsyncEngineArgs,
engine_config: Optional[VllmConfig] = None,
start_engine_loop: bool = True,
usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
stat_loggers: Optional[Dict[str, StatLoggerBase]] = None,
) -> "AsyncLLMEngine":
"""Creates an async LLM engine from the engine arguments."""
# Create the engine configs.
if engine_config is None:
engine_config = engine_args.create_engine_config(usage_context)
executor_class = cls._get_executor_cls(engine_config)
vllm_config = engine_args.create_engine_config(usage_context)
# Create the async LLM engine.
engine = cls(
vllm_config=engine_config,
executor_class=executor_class,
log_requests=not engine_args.disable_log_requests,
log_stats=not engine_args.disable_log_stats,
async_engine_cls = cls
if envs.VLLM_USE_V1:
from vllm.v1.engine.async_llm import AsyncLLM as V1AsyncLLMEngine
async_engine_cls = V1AsyncLLMEngine
return async_engine_cls.from_vllm_config(
vllm_config=vllm_config,
start_engine_loop=start_engine_loop,
usage_context=usage_context,
stat_loggers=stat_loggers,
disable_log_stats=engine_args.disable_log_stats,
disable_log_requests=engine_args.disable_log_requests,
)
return engine
@property
def is_running(self) -> bool:
@@ -1203,7 +1230,7 @@ class AsyncLLMEngine(EngineClient):
# TODO(v1): Remove this class proxy when V1 goes default.
if envs.VLLM_USE_V1:
if envs.is_set("VLLM_USE_V1") and envs.VLLM_USE_V1:
from vllm.v1.engine.async_llm import AsyncLLM
AsyncLLMEngine = AsyncLLM # type: ignore