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

47
vllm/engine/llm_engine.py
View File

@@ -216,6 +216,12 @@ class LLMEngine:
mm_registry: MultiModalRegistry = MULTIMODAL_REGISTRY,
use_cached_outputs: bool = False,
) -> None:
if envs.VLLM_USE_V1:
raise ValueError(
"Using V0 LLMEngine, but envs.VLLM_USE_V1=True. "
"This should not happen. As a workaround, try using "
"LLMEngine.from_vllm_config(...) or explicitly set "
"VLLM_USE_V1=0 or 1 and report this issue on Github.")
self.vllm_config = vllm_config
self.model_config = vllm_config.model_config
@@ -479,6 +485,22 @@ class LLMEngine:
f"{distributed_executor_backend}")
return executor_class
@classmethod
def from_vllm_config(
cls,
vllm_config: VllmConfig,
usage_context: UsageContext = UsageContext.ENGINE_CONTEXT,
stat_loggers: Optional[Dict[str, StatLoggerBase]] = None,
disable_log_stats: bool = False,
) -> "LLMEngine":
return cls(
vllm_config=vllm_config,
executor_class=cls._get_executor_cls(vllm_config),
log_stats=(not disable_log_stats),
usage_context=usage_context,
stat_loggers=stat_loggers,
)
@classmethod
def from_engine_args(
cls,
@@ -488,19 +510,20 @@ class LLMEngine:
) -> "LLMEngine":
"""Creates an LLM engine from the engine arguments."""
# Create the engine configs.
engine_config = engine_args.create_engine_config(usage_context)
executor_class = cls._get_executor_cls(engine_config)
# Create the LLM engine.
engine = cls(
vllm_config=engine_config,
executor_class=executor_class,
log_stats=not engine_args.disable_log_stats,
vllm_config = engine_args.create_engine_config(usage_context)
engine_cls = cls
if envs.VLLM_USE_V1:
from vllm.v1.engine.llm_engine import LLMEngine as V1LLMEngine
engine_cls = V1LLMEngine
return engine_cls.from_vllm_config(
vllm_config=vllm_config,
usage_context=usage_context,
stat_loggers=stat_loggers,
disable_log_stats=engine_args.disable_log_stats,
)
return engine
def __reduce__(self):
# This is to ensure that the LLMEngine is not referenced in
# the closure used to initialize Ray worker actors
@@ -2097,6 +2120,6 @@ class LLMEngine:
return sampling_params
# TODO(v1): Remove this class proxy when V1 goes default.
if envs.VLLM_USE_V1:
from vllm.v1.engine.llm_engine import LLMEngine # type: ignore
if envs.is_set("VLLM_USE_V1") and envs.VLLM_USE_V1:
from vllm.v1.engine.llm_engine import LLMEngine as V1LLMEngine
LLMEngine = V1LLMEngine # type: ignore