diff --git a/vllm/utils/argparse_utils.py b/vllm/utils/argparse_utils.py index 48f7ad9199..fc906cf227 100644 --- a/vllm/utils/argparse_utils.py +++ b/vllm/utils/argparse_utils.py @@ -16,7 +16,7 @@ from argparse import ( _ArgumentGroup, ) from collections import defaultdict -from typing import Any +from typing import Any, NoReturn import regex as re import yaml @@ -132,6 +132,22 @@ class FlexibleArgumentParser(ArgumentParser): # Pop kwarg "add_json_tip" to control whether to add the JSON tip self.add_json_tip = kwargs.pop("add_json_tip", True) super().__init__(*args, **kwargs) + self._show_serve_task_hint = False + + def error(self, message: str) -> NoReturn: + if ( + (self.prog.endswith(" serve") or self._show_serve_task_hint) + and message.startswith("unrecognized arguments:") + and re.search(r"(^|\s)--task(=|\s|$)", message) + ): + message += ( + "\n\nHint: --task is not a vllm serve option. " + "For embedding, reranking, or reward models, use " + "--runner pooling. To adapt a generative model for pooling, " + "also use --convert embed, --convert classify, or " + "--convert reward." + ) + super().error(message) if sys.version_info < (3, 13): # Enable the deprecated kwarg for Python 3.12 and below @@ -247,6 +263,9 @@ class FlexibleArgumentParser(ArgumentParser): ): if args is None: args = sys.argv[1:] + self._show_serve_task_hint = args[:1] == ["serve"] and any( + re.match(r"^--task(=.+|$)", arg) for arg in args[1:] + ) if args and args[0] == "serve": # Check for --model in command line arguments first