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, 10:45 UTC
Cap the number of API servers to 1 when using Elastic EP. (#37466)
Signed-off-by: Sage Moore <sage@neuralmagic.com>
Sage Moore committed
Mar 19, 2026 at 07:42 UTC
7c0cf3bcd0867b5420e6ad4f2ff6b2b25e73b022
1 file changed
+9
vllm/entrypoints/cli/serve.py
+9
@@ -108,6 +108,15 @@ class ServeSubcommand(CLISubcommand):
108
args.api_server_count,
109
)
110
111
+ # Elastic EP currently only supports running with at most one API server.
112
+ if getattr(args, "enable_elastic_ep", False) and args.api_server_count > 1:
113
+ logger.warning(
114
+ "Elastic EP only supports running with with at most one API server. "
115
+ "Capping api_server_count from %d to 1.",
116
+ args.api_server_count,
117
+ )
118
+ args.api_server_count = 1
119
+
120
if args.api_server_count < 1:
121
run_headless(args)
122
elif args.api_server_count > 1: