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

[Core] Store only the keys for multi-modal data in P0 (#22198)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-08-07 16:45:04 +08:00
committed by GitHub
parent 289b18e670
commit 766bc8162c
17 changed files with 325 additions and 234 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
vllm/envs.py
View File

@@ -65,7 +65,7 @@ if TYPE_CHECKING:
VLLM_AUDIO_FETCH_TIMEOUT: int = 10
VLLM_MAX_AUDIO_CLIP_FILESIZE_MB: int = 25
VLLM_VIDEO_LOADER_BACKEND: str = "opencv"
VLLM_MM_INPUT_CACHE_GIB: int = 8
VLLM_MM_INPUT_CACHE_GIB: int = 4
VLLM_TARGET_DEVICE: str = "cuda"
MAX_JOBS: Optional[str] = None
NVCC_THREADS: Optional[str] = None
@@ -561,8 +561,8 @@ environment_variables: dict[str, Callable[[], Any]] = {
"VLLM_VIDEO_LOADER_BACKEND":
lambda: os.getenv("VLLM_VIDEO_LOADER_BACKEND", "opencv"),
# Cache size (in GiB) for multimodal input cache
# Default is 4 GiB
# Cache size (in GiB per process) for multimodal input cache
# Default is 4 GiB per API process + 4 GiB per engine core process
"VLLM_MM_INPUT_CACHE_GIB":
lambda: int(os.getenv("VLLM_MM_INPUT_CACHE_GIB", "4")),