[Misc] Configurable timeout for execute_model RPC calls via env var (#19544)

Signed-off-by: jinqinn <goodqinjin@163.com>
This commit is contained in:
jinqinn
2025-06-22 20:36:26 -07:00
committed by GitHub
parent 4a0f7888a3
commit f39ab2d4bd
2 changed files with 12 additions and 11 deletions
+6
View File
@@ -130,6 +130,7 @@ if TYPE_CHECKING:
VLLM_TOOL_PARSE_REGEX_TIMEOUT_SECONDS: int = 1
VLLM_SLEEP_WHEN_IDLE: bool = False
VLLM_MQ_MAX_CHUNK_BYTES_MB: int = 16
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: int = 300
VLLM_KV_CACHE_LAYOUT: Optional[str] = None
VLLM_COMPUTE_NANS_IN_LOGITS: bool = False
@@ -897,6 +898,11 @@ environment_variables: dict[str, Callable[[], Any]] = {
"VLLM_MQ_MAX_CHUNK_BYTES_MB":
lambda: int(os.getenv("VLLM_MQ_MAX_CHUNK_BYTES_MB", "16")),
# Timeout in seconds for execute_model RPC calls in multiprocessing
# executor (only applies when TP > 1).
"VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS":
lambda: int(os.getenv("VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS", "300")),
# KV Cache layout used throughout vllm.
# Some common values are:
# - NHD