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 6, 21:45 UTC
[Misc] Use envs module to get VLLM_DISABLED_KERNELS (#35776)
Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
Martin Hickey committed
Mar 11, 2026 at 13:37 UTC
700a1ddc65dfbf3590ff746013cd4070fb41c01d
1 file changed
+1
-3
vllm/model_executor/kernels/linear/__init__.py
+1
-3
@@ -13,7 +13,6 @@ or kernel implementation, add it to this __init__.py to maintain
13
import stability.
14
"""
15
16
-import os
16
from typing import TypeVar
17
18
import torch
@@ -154,8 +153,7 @@ _KernelConfigT = TypeVar("_KernelConfigT", bound=ScaledMMLinearLayerConfig)
153
def is_supported_and_can_implement_kernel(
154
kernel: type[_KernelT], config: _KernelConfigT, compute_capability: int | None
155
) -> tuple[bool, str]:
157
- # TODO: Fetch `VLLM_DISABLED_KERNELS` from vllm.envs instead.
158
- if kernel.__name__ in os.environ.get("VLLM_DISABLED_KERNELS", "").split(","):
156
+ if kernel.__name__ in envs.VLLM_DISABLED_KERNELS:
157
return False, f" {kernel.__name__} is disabled by environment variable"
158
159
if compute_capability is None: