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
[BugFix] Fix DP headless mode arg validation (#20398)
Signed-off-by: Nick Hill <nhill@redhat.com>
Nick Hill committed
Jul 3, 2025 at 01:15 UTC
059d4cdb4907025b27f1326b47566224a1487b06
1 file changed
+4
-4
vllm/entrypoints/cli/serve.py
+4
-4
@@ -119,6 +119,10 @@ def run_headless(args: argparse.Namespace):
119
if not envs.VLLM_USE_V1:
120
raise ValueError("Headless mode is only supported for V1")
121
122
+ if engine_args.data_parallel_rank is not None:
123
+ raise ValueError("data_parallel_rank is not applicable in "
124
+ "headless mode")
125
+
126
parallel_config = vllm_config.parallel_config
127
local_engine_count = parallel_config.data_parallel_size_local
128
@@ -126,10 +130,6 @@ def run_headless(args: argparse.Namespace):
130
raise ValueError("data_parallel_size_local must be > 0 in "
131
"headless mode")
132
129
- if parallel_config.data_parallel_rank is not None:
130
- raise ValueError("data_parallel_rank is not applicable in "
131
- "headless mode")
132
-
133
host = parallel_config.data_parallel_master_ip
134
port = engine_args.data_parallel_rpc_port # add to config too
135
handshake_address = get_tcp_uri(host, port)