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 14, 05:30 UTC
Cache the environment variable check for batch invariance (#26510)
Signed-off-by: Bram Wasti <bwasti@meta.com>
Bram Wasti committed
Oct 10, 2025 at 15:47 UTC
e317414ce11a66f6e653ce382f5d987b0c0191bd
1 file changed
+6
-3
csrc/core/batch_invariant.hpp
+6
-3
@@ -8,9 +8,12 @@ namespace vllm {
8
// vllm_kernel_override_batch_invariant(); returns true
9
// if env VLLM_KERNEL_OVERRIDE_BATCH_INVARIANT=1
10
inline bool vllm_kernel_override_batch_invariant() {
11
- std::string env_key = "VLLM_KERNEL_OVERRIDE_BATCH_INVARIANT";
12
- const char* val = std::getenv(env_key.c_str());
13
- return (val && std::atoi(val) != 0) ? 1 : 0;
11
+ static bool cached = []() {
12
+ std::string env_key = "VLLM_KERNEL_OVERRIDE_BATCH_INVARIANT";
13
+ const char* val = std::getenv(env_key.c_str());
14
+ return (val && std::atoi(val) != 0) ? 1 : 0;
15
+ }();
16
+ return cached;
17
}
18
19
} // namespace vllm