xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-02 05:00:53 +08:00
Code Issues Packages Projects Releases Wiki Activity

Disable outlines cache by default (#14837)

Browse Source
This commit is contained in:
Russell Bryant 2025-03-14 23:57:55 -04:00 committed by GitHub
parent ccf02fcbae
commit 776dcec8fe
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 16 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
vllm/envs.py
View File

@ -95,6 +95,7 @@ if TYPE_CHECKING:
VLLM_DP_MASTER_IP: str = ""
VLLM_DP_MASTER_PORT: int = 0
VLLM_MARLIN_USE_ATOMIC_ADD: bool = False
VLLM_V0_USE_OUTLINES_CACHE: bool = False
def get_default_cache_root():
@ -623,6 +624,12 @@ environment_variables: dict[str, Callable[[], Any]] = {
# Whether to use atomicAdd reduce in gptq/awq marlin kernel.
"VLLM_MARLIN_USE_ATOMIC_ADD":
lambda: os.environ.get("VLLM_MARLIN_USE_ATOMIC_ADD", "0") == "1",
# Whether to turn on the outlines cache for V0
# This cache is unbounded and on disk, so it's not safe to use in
# an environment with potentially malicious users.
"VLLM_V0_USE_OUTLINES_CACHE":
lambda: os.environ.get("VLLM_V0_USE_OUTLINES_CACHE", "0") == "1",
}
# end-env-vars-definition

10
vllm/model_executor/guided_decoding/outlines_logits_processors.py
View File

@ -24,7 +24,7 @@ from typing import Callable, DefaultDict, Dict, List, Optional, Union
import numpy as np
import torch
from outlines import grammars
from outlines.caching import cache
from outlines.caching import cache, disable_cache
from outlines.fsm.guide import (CFGGuide, CFGState, Generate, Guide,
RegexGuide, Write)
from outlines.fsm.parsing import PartialLark
@ -32,12 +32,20 @@ from outlines_core.fsm.json_schema import build_regex_from_schema
from pydantic import BaseModel
from transformers import PreTrainedTokenizerBase
import vllm.envs as envs
from vllm.logger import init_logger
from vllm.model_executor.guided_decoding.reasoner import Reasoner
from vllm.platforms import current_platform
logger = init_logger(__name__)
if envs.VLLM_V0_USE_OUTLINES_CACHE:
logger.warning("Enabling outlines cache. This is an unbounded on-disk "
"cache. It may consume a lot of disk space and should "
"not be used with untrusted clients.")
else:
disable_cache()
class BaseLogitsProcessor: