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/v1/engine/async_llm.py
View File

@@ -8,6 +8,7 @@ from typing import Optional, Union
import numpy as np
import vllm.envs as envs
from vllm.config import ModelConfig, VllmConfig
from vllm.engine.arg_utils import AsyncEngineArgs
from vllm.engine.protocol import EngineClient
@@ -49,6 +50,12 @@ class AsyncLLM(EngineClient):
log_requests: bool = True,
start_engine_loop: bool = True,
) -> None:
if not envs.VLLM_USE_V1:
raise ValueError(
"Using V1 AsyncLLMEngine, but envs.VLLM_USE_V1=False. "
"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.")
assert start_engine_loop
@@ -92,22 +99,50 @@ class AsyncLLM(EngineClient):
self.output_handler: Optional[asyncio.Task] = None
@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,
) -> "AsyncLLM":
if not envs.VLLM_USE_V1:
raise ValueError(
"Using V1 AsyncLLMEngine, but envs.VLLM_USE_V1=False. "
"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.")
# FIXME(rob): refactor VllmConfig to include the StatLoggers
# include StatLogger in the Oracle decision.
if stat_loggers is not None:
raise ValueError("Custom StatLoggers are not yet supported on V1. "
"Explicitly set VLLM_USE_V1=0 to disable V1.")
# Create the LLMEngine.
return cls(
vllm_config=vllm_config,
executor_class=Executor.get_class(vllm_config),
start_engine_loop=start_engine_loop,
log_requests=not disable_log_requests,
log_stats=not disable_log_stats,
usage_context=usage_context,
)
@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,
) -> "AsyncLLM":
"""Create an AsyncLLM from the EngineArgs."""
# Create the engine configs.
if engine_config is None:
vllm_config = engine_args.create_engine_config(usage_context)
else:
vllm_config = engine_config
vllm_config = engine_args.create_engine_config(usage_context)
executor_class = Executor.get_class(vllm_config)
# Create the AsyncLLM.