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

[Core] V1: Use multiprocessing by default (#11074)

Browse Source 
Signed-off-by: Russell Bryant <rbryant@redhat.com>
This commit is contained in:
Russell Bryant
2024-12-13 19:27:32 -05:00
committed by GitHub
parent 0d8451c3a4
commit 4863e5fba5
10 changed files with 299 additions and 17 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/envs.py
View File

@@ -69,7 +69,7 @@ if TYPE_CHECKING:
VLLM_SKIP_P2P_CHECK: bool = False
VLLM_DISABLED_KERNELS: List[str] = []
VLLM_USE_V1: bool = False
VLLM_ENABLE_V1_MULTIPROCESSING: bool = False
VLLM_ENABLE_V1_MULTIPROCESSING: bool = True
VLLM_LOG_BATCHSIZE_INTERVAL: float = -1
@@ -460,7 +460,7 @@ environment_variables: Dict[str, Callable[[], Any]] = {
# If set, enable multiprocessing in LLM for the V1 code path.
"VLLM_ENABLE_V1_MULTIPROCESSING":
lambda: bool(int(os.getenv("VLLM_ENABLE_V1_MULTIPROCESSING", "0"))),
lambda: bool(int(os.getenv("VLLM_ENABLE_V1_MULTIPROCESSING", "1"))),
"VLLM_LOG_BATCHSIZE_INTERVAL":
lambda: float(os.getenv("VLLM_LOG_BATCHSIZE_INTERVAL", "-1")),
}