A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 13, 17:00 UTC
[Bug] Fix env string `"0"` same to `True` (#28159)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Nov 5, 2025 at 20:04 UTC
90189c71a9629cf2c866b213d6d28b08937a7566
1 file changed
+6
-4
vllm/envs.py
+6
-4
@@ -776,7 +776,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
776
# If set, the OpenAI API server will stay alive even after the underlying
777
# AsyncLLMEngine errors and stops serving requests
778
"VLLM_KEEP_ALIVE_ON_ENGINE_DEATH": lambda: bool(
779
- os.getenv("VLLM_KEEP_ALIVE_ON_ENGINE_DEATH", 0)
779
+ int(os.getenv("VLLM_KEEP_ALIVE_ON_ENGINE_DEATH", "0"))
780
),
781
# If the env var VLLM_ALLOW_LONG_MAX_MODEL_LEN is set, it allows
782
# the user to specify a max sequence length greater than
@@ -1313,7 +1313,9 @@ environment_variables: dict[str, Callable[[], Any]] = {
1313
),
1314
# If set, it means we pre-downloaded cubin files and flashinfer will
1315
# read the cubin files directly.
1316
- "VLLM_HAS_FLASHINFER_CUBIN": lambda: os.getenv("VLLM_HAS_FLASHINFER_CUBIN", False),
1316
+ "VLLM_HAS_FLASHINFER_CUBIN": lambda: bool(
1317
+ int(os.getenv("VLLM_HAS_FLASHINFER_CUBIN", "0"))
1318
+ ),
1319
# Supported options:
1320
# - "flashinfer-cudnn": use flashinfer cudnn GEMM backend
1321
# - "flashinfer-trtllm": use flashinfer trtllm GEMM backend
@@ -1449,8 +1451,8 @@ environment_variables: dict[str, Callable[[], Any]] = {
1451
# top 5 collected objects
1452
"VLLM_GC_DEBUG": lambda: os.getenv("VLLM_GC_DEBUG", ""),
1453
# Disables parallel execution of shared_experts via separate cuda stream
1452
- "VLLM_DISABLE_SHARED_EXPERTS_STREAM": lambda: os.getenv(
1453
- "VLLM_DISABLE_SHARED_EXPERTS_STREAM", False
1454
+ "VLLM_DISABLE_SHARED_EXPERTS_STREAM": lambda: bool(
1455
+ int(os.getenv("VLLM_DISABLE_SHARED_EXPERTS_STREAM", "0"))
1456
),
1457
# Format for saving torch.compile cache artifacts
1458
# - "binary": saves as binary file