cli_args.py 6.02 KB
Newer Older
1
2
3
4
5
6
7
8
9
10
"""
This file contains the command line arguments for the vLLM's
OpenAI-compatible server. It is kept in a separate file for documentation
purposes.
"""

import argparse
import json
import ssl

11
from vllm.engine.arg_utils import AsyncEngineArgs, nullable_str
12
13
from vllm.entrypoints.openai.serving_engine import (LoRAModulePath,
                                                    PromptAdapterPath)
14
from vllm.utils import FlexibleArgumentParser
15
16
17
18
19
20
21
22


class LoRAParserAction(argparse.Action):

    def __call__(self, parser, namespace, values, option_string=None):
        lora_list = []
        for item in values:
            name, path = item.split('=')
23
            lora_list.append(LoRAModulePath(name, path))
24
25
26
        setattr(namespace, self.dest, lora_list)


27
28
29
30
31
32
33
34
35
36
class PromptAdapterParserAction(argparse.Action):

    def __call__(self, parser, namespace, values, option_string=None):
        adapter_list = []
        for item in values:
            name, path = item.split('=')
            adapter_list.append(PromptAdapterPath(name, path))
        setattr(namespace, self.dest, adapter_list)


Ethan Xu's avatar
Ethan Xu committed
37
def make_arg_parser(parser: FlexibleArgumentParser) -> FlexibleArgumentParser:
38
39
40
41
    parser.add_argument("--host",
                        type=nullable_str,
                        default=None,
                        help="host name")
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
    parser.add_argument("--port", type=int, default=8000, help="port number")
    parser.add_argument(
        "--uvicorn-log-level",
        type=str,
        default="info",
        choices=['debug', 'info', 'warning', 'error', 'critical', 'trace'],
        help="log level for uvicorn")
    parser.add_argument("--allow-credentials",
                        action="store_true",
                        help="allow credentials")
    parser.add_argument("--allowed-origins",
                        type=json.loads,
                        default=["*"],
                        help="allowed origins")
    parser.add_argument("--allowed-methods",
                        type=json.loads,
                        default=["*"],
                        help="allowed methods")
    parser.add_argument("--allowed-headers",
                        type=json.loads,
                        default=["*"],
                        help="allowed headers")
    parser.add_argument("--api-key",
65
                        type=nullable_str,
66
67
68
69
70
                        default=None,
                        help="If provided, the server will require this key "
                        "to be presented in the header.")
    parser.add_argument(
        "--lora-modules",
71
        type=nullable_str,
72
73
74
75
76
        default=None,
        nargs='+',
        action=LoRAParserAction,
        help="LoRA module configurations in the format name=path. "
        "Multiple modules can be specified.")
77
78
79
80
81
82
83
84
    parser.add_argument(
        "--prompt-adapters",
        type=nullable_str,
        default=None,
        nargs='+',
        action=PromptAdapterParserAction,
        help="Prompt adapter configurations in the format name=path. "
        "Multiple adapters can be specified.")
85
    parser.add_argument("--chat-template",
86
                        type=nullable_str,
87
88
89
90
91
                        default=None,
                        help="The file path to the chat template, "
                        "or the template in single-line form "
                        "for the specified model")
    parser.add_argument("--response-role",
92
                        type=nullable_str,
93
94
95
96
                        default="assistant",
                        help="The role name to return if "
                        "`request.add_generation_prompt=true`.")
    parser.add_argument("--ssl-keyfile",
97
                        type=nullable_str,
98
99
100
                        default=None,
                        help="The file path to the SSL key file")
    parser.add_argument("--ssl-certfile",
101
                        type=nullable_str,
102
103
104
                        default=None,
                        help="The file path to the SSL cert file")
    parser.add_argument("--ssl-ca-certs",
105
                        type=nullable_str,
106
107
108
109
110
111
112
113
114
115
                        default=None,
                        help="The CA certificates file")
    parser.add_argument(
        "--ssl-cert-reqs",
        type=int,
        default=int(ssl.CERT_NONE),
        help="Whether client certificate is required (see stdlib ssl module's)"
    )
    parser.add_argument(
        "--root-path",
116
        type=nullable_str,
117
118
119
120
        default=None,
        help="FastAPI root_path when app is behind a path based routing proxy")
    parser.add_argument(
        "--middleware",
121
        type=nullable_str,
122
123
124
125
126
127
128
129
130
        action="append",
        default=[],
        help="Additional ASGI middleware to apply to the app. "
        "We accept multiple --middleware arguments. "
        "The value should be an import path. "
        "If a function is provided, vLLM will add it to the server "
        "using @app.middleware('http'). "
        "If a class is provided, vLLM will add it to the server "
        "using app.add_middleware(). ")
131
132
133
    parser.add_argument(
        "--return-tokens-as-token-ids",
        action="store_true",
134
135
        help="When --max-logprobs is specified, represents single tokens as "
        "strings of the form 'token_id:{token_id}' so that tokens that "
136
        "are not JSON-encodable can be identified.")
137
138
139
140
141
    parser.add_argument(
        "--disable-frontend-multiprocessing",
        action="store_true",
        help="If specified, will run the OpenAI frontend server in the same "
        "process as the model serving engine.")
142
143

    parser = AsyncEngineArgs.add_cli_args(parser)
144
145
146
147
148
149
150
151

    parser.add_argument('--max-log-len',
                        type=int,
                        default=None,
                        help='Max number of prompt characters or prompt '
                        'ID numbers being printed in log.'
                        '\n\nDefault: Unlimited')

152
    return parser
Ethan Xu's avatar
Ethan Xu committed
153
154
155
156
157
158


def create_parser_for_docs() -> FlexibleArgumentParser:
    parser_for_docs = FlexibleArgumentParser(
        prog="-m vllm.entrypoints.openai.api_server")
    return make_arg_parser(parser_for_docs)