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 15, 03:30 UTC
[Misc] improve error msg (#20604)
Signed-off-by: reidliu41 <reid201711@gmail.com>
Reid committed
Jul 8, 2025 at 12:45 UTC
8369b7c2a9524681b377085811bc176e332594c6
1 file changed
+6
-3
vllm/entrypoints/cli/serve.py
+6
-3
@@ -46,8 +46,10 @@ class ServeSubcommand(CLISubcommand):
46
run_headless(args)
47
else:
48
if args.data_parallel_start_rank:
49
- raise ValueError("data_parallel_start_rank is only "
50
- "applicable in headless mode")
49
+ raise ValueError(
50
+ "data_parallel_start_rank is only applicable "
51
+ "in headless mode. "
52
+ "Add --headless flag to enable headless mode.")
53
if args.api_server_count > 1:
54
run_multi_api_server(args)
55
else:
@@ -81,7 +83,8 @@ class ServeSubcommand(CLISubcommand):
83
'-dpr',
84
type=int,
85
default=0,
84
- help='Starting data parallel rank for secondary nodes.')
86
+ help="Starting data parallel rank for secondary nodes. "
87
+ "Requires --headless.")
88
serve_parser.add_argument('--api-server-count',
89
'-asc',
90
type=int,