Propagate Flash Attention cache configuration to Ray workers (#49177)
Signed-off-by: Stefan Kaestle <skaestle@nvidia.com>
Stefan Kaestle committed
Jul 21, 2026 at 09:47 UTC
adfbbc10051fff393b3f6a3b0e96f8d6243c8cb9
1 file changed
+1
vllm/ray/ray_env.py
+1
@@ -35,6 +35,7 @@ except json.JSONDecodeError:
35
# ---------------------------------------------------------------------------
36
DEFAULT_ENV_VAR_PREFIXES: set[str] = {
37
"VLLM_",
38
+ "FLASH_ATTENTION_",
39
"LMCACHE_",
40
"NCCL_",
41
"UCX_",