[Frontend] improve vllm run-batch --help display (#19187)

Signed-off-by: reidliu41 <reid201711@gmail.com>
Co-authored-by: reidliu41 <reid201711@gmail.com>

vllm/entrypoints/cli/main.py

@@ -11,7 +11,7 @@ import vllm.entrypoints.cli.openai
 import vllm.entrypoints.cli.run_batch
 import vllm.entrypoints.cli.serve
 import vllm.version
-from vllm.entrypoints.utils import VLLM_SERVE_PARSER_EPILOG, cli_env_setup
+from vllm.entrypoints.utils import VLLM_SUBCMD_PARSER_EPILOG, cli_env_setup
 from vllm.utils import FlexibleArgumentParser
 
 CMD_MODULES = [
@@ -37,7 +37,7 @@ def main():
 
     parser = FlexibleArgumentParser(
         description="vLLM CLI",
-        epilog=VLLM_SERVE_PARSER_EPILOG,
+        epilog=VLLM_SUBCMD_PARSER_EPILOG,
     )
     parser.add_argument('-v',
                         '--version',

vllm/entrypoints/cli/run_batch.py

@@ -10,6 +10,8 @@ from vllm.entrypoints.cli.types import CLISubcommand
 from vllm.entrypoints.logger import logger
 from vllm.entrypoints.openai.run_batch import main as run_batch_main
 from vllm.entrypoints.openai.run_batch import make_arg_parser
+from vllm.entrypoints.utils import (VLLM_SUBCMD_PARSER_EPILOG,
+                                    show_filtered_argument_or_group_from_help)
 from vllm.utils import FlexibleArgumentParser
 from vllm.version import __version__ as VLLM_VERSION
 
@@ -49,7 +51,11 @@ class RunBatchSubcommand(CLISubcommand):
             usage=
             "vllm run-batch -i INPUT.jsonl -o OUTPUT.jsonl --model <model>",
         )
-        return make_arg_parser(run_batch_parser)
+        run_batch_parser = make_arg_parser(run_batch_parser)
+        show_filtered_argument_or_group_from_help(run_batch_parser,
+                                                  "run-batch")
+        run_batch_parser.epilog = VLLM_SUBCMD_PARSER_EPILOG
+        return run_batch_parser
 
 
 def cmd_init() -> list[CLISubcommand]:

vllm/entrypoints/cli/serve.py

@@ -16,7 +16,7 @@ from vllm.entrypoints.openai.api_server import (run_server, run_server_worker,
                                                 setup_server)
 from vllm.entrypoints.openai.cli_args import (make_arg_parser,
                                               validate_parsed_serve_args)
-from vllm.entrypoints.utils import (VLLM_SERVE_PARSER_EPILOG,
+from vllm.entrypoints.utils import (VLLM_SUBCMD_PARSER_EPILOG,
                                     show_filtered_argument_or_group_from_help)
 from vllm.executor.multiproc_worker_utils import _add_prefix
 from vllm.logger import init_logger
@@ -101,8 +101,8 @@ class ServeSubcommand(CLISubcommand):
         )
 
         serve_parser = make_arg_parser(serve_parser)
-        show_filtered_argument_or_group_from_help(serve_parser)
-        serve_parser.epilog = VLLM_SERVE_PARSER_EPILOG
+        show_filtered_argument_or_group_from_help(serve_parser, "serve")
+        serve_parser.epilog = VLLM_SUBCMD_PARSER_EPILOG
         return serve_parser
 
 

vllm/entrypoints/utils.py

@@ -14,8 +14,9 @@ from vllm.logger import init_logger
 
 logger = init_logger(__name__)
 
-VLLM_SERVE_PARSER_EPILOG = (
-    "Tip: Use `vllm serve --help=<keyword>` to explore arguments from help.\n"
+VLLM_SUBCMD_PARSER_EPILOG = (
+    "Tip: Use `vllm [serve|run-batch] --help=<keyword>` "
+    "to explore arguments from help.\n"
     "   - To view a argument group:     --help=ModelConfig\n"
     "   - To view a single argument:    --help=max-num-seqs\n"
     "   - To search by keyword:         --help=max\n"
@@ -173,8 +174,15 @@ def _validate_truncation_size(
     return truncate_prompt_tokens
 
 
-def show_filtered_argument_or_group_from_help(parser):
+def show_filtered_argument_or_group_from_help(parser, subcommand_name):
     import sys
+
+    # Only handle --help=<keyword> for the current subcommand.
+    # Since subparser_init() runs for all subcommands during CLI setup,
+    # we skip processing if the subcommand name is not in sys.argv.
+    if subcommand_name not in sys.argv:
+        return
+
     for arg in sys.argv:
         if arg.startswith('--help='):
             search_keyword = arg.split('=', 1)[1]