test_arg_utils.py 11.8 KB
Newer Older
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3

4
import json
5
from argparse import ArgumentError, ArgumentTypeError
6
7
from contextlib import nullcontext
from dataclasses import dataclass, field
8
from typing import Annotated, Literal, Optional
9

10
11
import pytest

12
from vllm.config import CompilationConfig, config
13
from vllm.engine.arg_utils import (EngineArgs, contains_type, get_kwargs,
14
15
                                   get_type, get_type_hints, is_not_builtin,
                                   is_type, literal_to_kwargs, nullable_kvs,
16
                                   optional_type, parse_type)
17
18
19
from vllm.utils import FlexibleArgumentParser


20
21
22
23
24
25
26
27
28
29
30
31
32
@pytest.mark.parametrize(("type", "value", "expected"), [
    (int, "42", 42),
    (float, "3.14", 3.14),
    (str, "Hello World!", "Hello World!"),
    (json.loads, '{"foo":1,"bar":2}', {
        "foo": 1,
        "bar": 2
    }),
    (json.loads, "foo=1,bar=2", {
        "foo": 1,
        "bar": 2
    }),
])
33
34
def test_parse_type(type, value, expected):
    parse_type_func = parse_type(type)
35
36
37
38
    context = nullcontext()
    if value == "foo=1,bar=2":
        context = pytest.warns(DeprecationWarning)
    with context:
39
40
41
42
43
44
45
        assert parse_type_func(value) == expected


def test_optional_type():
    optional_type_func = optional_type(int)
    assert optional_type_func("None") is None
    assert optional_type_func("42") == 42
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77


@pytest.mark.parametrize(("type_hint", "type", "expected"), [
    (int, int, True),
    (int, float, False),
    (list[int], list, True),
    (list[int], tuple, False),
    (Literal[0, 1], Literal, True),
])
def test_is_type(type_hint, type, expected):
    assert is_type(type_hint, type) == expected


@pytest.mark.parametrize(("type_hints", "type", "expected"), [
    ({float, int}, int, True),
    ({int, tuple[int]}, int, True),
    ({int, tuple[int]}, float, False),
    ({str, Literal["x", "y"]}, Literal, True),
])
def test_contains_type(type_hints, type, expected):
    assert contains_type(type_hints, type) == expected


@pytest.mark.parametrize(("type_hints", "type", "expected"), [
    ({int, float}, int, int),
    ({int, float}, str, None),
    ({str, Literal["x", "y"]}, Literal, Literal["x", "y"]),
])
def test_get_type(type_hints, type, expected):
    assert get_type(type_hints, type) == expected


78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
@pytest.mark.parametrize(("type_hints", "expected"), [
    ({Literal[1, 2]}, {
        "type": int,
        "choices": [1, 2]
    }),
    ({Literal[1, "a"]}, Exception),
])
def test_literal_to_kwargs(type_hints, expected):
    context = nullcontext()
    if expected is Exception:
        context = pytest.raises(expected)
    with context:
        assert literal_to_kwargs(type_hints) == expected


93
94
@config
@dataclass
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
class NestedConfig:
    field: int = 1
    """field"""


@config
@dataclass
class FromCliConfig1:
    field: int = 1
    """field"""

    @classmethod
    def from_cli(cls, cli_value: str):
        inst = cls(**json.loads(cli_value))
        inst.field += 1
        return inst


@config
@dataclass
class FromCliConfig2:
    field: int = 1
    """field"""

    @classmethod
    def from_cli(cls, cli_value: str):
        inst = cls(**json.loads(cli_value))
        inst.field += 2
        return inst


@config
@dataclass
class DummyConfig:
129
130
131
132
133
134
135
    regular_bool: bool = True
    """Regular bool with default True"""
    optional_bool: Optional[bool] = None
    """Optional bool with default None"""
    optional_literal: Optional[Literal["x", "y"]] = None
    """Optional literal with default None"""
    tuple_n: tuple[int, ...] = field(default_factory=lambda: (1, 2, 3))
136
    """Tuple with variable length"""
137
    tuple_2: tuple[int, int] = field(default_factory=lambda: (1, 2))
138
    """Tuple with fixed length"""
139
    list_n: list[int] = field(default_factory=lambda: [1, 2, 3])
140
141
142
143
144
    """List with variable length"""
    list_literal: list[Literal[1, 2]] = field(default_factory=list)
    """List with literal choices"""
    literal_literal: Literal[Literal[1], Literal[2]] = 1
    """Literal of literals with default 1"""
145
146
    json_tip: dict = field(default_factory=dict)
    """Dict which will be JSON in CLI"""
147
148
149
150
151
152
    nested_config: NestedConfig = field(default_factory=NestedConfig)
    """Nested config"""
    from_cli_config1: FromCliConfig1 = field(default_factory=FromCliConfig1)
    """Config with from_cli method"""
    from_cli_config2: FromCliConfig2 = field(default_factory=FromCliConfig2)
    """Different config with from_cli method"""
153
154
155
156


@pytest.mark.parametrize(("type_hint", "expected"), [
    (int, False),
157
    (DummyConfig, True),
158
159
160
161
162
])
def test_is_not_builtin(type_hint, expected):
    assert is_not_builtin(type_hint) == expected


163
164
165
166
167
168
169
170
171
172
173
174
@pytest.mark.parametrize(
    ("type_hint", "expected"), [
        (Annotated[int, "annotation"], {int}),
        (Optional[int], {int, type(None)}),
        (Annotated[Optional[int], "annotation"], {int, type(None)}),
        (Optional[Annotated[int, "annotation"]], {int, type(None)}),
    ],
    ids=["Annotated", "Optional", "Annotated_Optional", "Optional_Annotated"])
def test_get_type_hints(type_hint, expected):
    assert get_type_hints(type_hint) == expected


175
def test_get_kwargs():
176
    kwargs = get_kwargs(DummyConfig)
177
178
179
180
181
182
183
184
185
186
187
188
189
    print(kwargs)

    # bools should not have their type set
    assert kwargs["regular_bool"].get("type") is None
    assert kwargs["optional_bool"].get("type") is None
    # optional literals should have None as a choice
    assert kwargs["optional_literal"]["choices"] == ["x", "y", "None"]
    # tuples should have the correct nargs
    assert kwargs["tuple_n"]["nargs"] == "+"
    assert kwargs["tuple_2"]["nargs"] == 2
    # lists should work
    assert kwargs["list_n"]["type"] is int
    assert kwargs["list_n"]["nargs"] == "+"
190
191
192
193
194
195
    # lists with literals should have the correct choices
    assert kwargs["list_literal"]["type"] is int
    assert kwargs["list_literal"]["nargs"] == "+"
    assert kwargs["list_literal"]["choices"] == [1, 2]
    # literals of literals should have merged choices
    assert kwargs["literal_literal"]["choices"] == [1, 2]
196
    # dict should have json tip in help
197
198
    json_tip = "Should either be a valid JSON string or JSON keys"
    assert json_tip in kwargs["json_tip"]["help"]
199
200
201
202
203
    # nested config should should construct the nested config
    assert kwargs["nested_config"]["type"]('{"field": 2}') == NestedConfig(2)
    # from_cli configs should be constructed with the correct method
    assert kwargs["from_cli_config1"]["type"]('{"field": 2}').field == 3
    assert kwargs["from_cli_config2"]["type"]('{"field": 2}').field == 4
204
205


206
@pytest.mark.parametrize(("arg", "expected"), [
207
    (None, dict()),
208
209
210
211
212
213
214
    ("image=16", {
        "image": 16
    }),
    ("image=16,video=2", {
        "image": 16,
        "video": 2
    }),
215
216
217
218
    ("Image=16, Video=2", {
        "image": 16,
        "video": 2
    }),
219
220
])
def test_limit_mm_per_prompt_parser(arg, expected):
221
222
223
224
    """This functionality is deprecated and will be removed in the future.
    This argument should be passed as JSON string instead.
    
    TODO: Remove with nullable_kvs."""
225
226
227
228
229
230
231
    parser = EngineArgs.add_cli_args(FlexibleArgumentParser())
    if arg is None:
        args = parser.parse_args([])
    else:
        args = parser.parse_args(["--limit-mm-per-prompt", arg])

    assert args.limit_mm_per_prompt == expected
232
233


234
235
236
237
238
def test_compilation_config():
    parser = EngineArgs.add_cli_args(FlexibleArgumentParser())

    # default value
    args = parser.parse_args([])
239
    assert args.compilation_config == CompilationConfig()
240
241

    # set to O3
242
243
    args = parser.parse_args(["-O0"])
    assert args.compilation_config.level == 0
244
245

    # set to O 3 (space)
246
247
    args = parser.parse_args(["-O", "1"])
    assert args.compilation_config.level == 1
248
249

    # set to O 3 (equals)
250
251
252
253
254
    args = parser.parse_args(["-O=2"])
    assert args.compilation_config.level == 2

    # set to O.level 3
    args = parser.parse_args(["-O.level", "3"])
255
256
    assert args.compilation_config.level == 3

257
    # set to string form of a dict
258
    args = parser.parse_args([
259
260
261
        "-O",
        '{"level": 3, "cudagraph_capture_sizes": [1, 2, 4, 8], '
        '"use_inductor": false}',
262
263
    ])
    assert (args.compilation_config.level == 3 and
264
265
            args.compilation_config.cudagraph_capture_sizes == [1, 2, 4, 8]
            and not args.compilation_config.use_inductor)
266

267
    # set to string form of a dict
268
269
    args = parser.parse_args([
        "--compilation-config="
270
271
        '{"level": 3, "cudagraph_capture_sizes": [1, 2, 4, 8], '
        '"use_inductor": true}',
272
273
    ])
    assert (args.compilation_config.level == 3 and
274
275
            args.compilation_config.cudagraph_capture_sizes == [1, 2, 4, 8]
            and args.compilation_config.use_inductor)
276
277


278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
def test_prefix_cache_default():
    parser = EngineArgs.add_cli_args(FlexibleArgumentParser())
    args = parser.parse_args([])

    engine_args = EngineArgs.from_cli_args(args=args)
    assert (not engine_args.enable_prefix_caching
            ), "prefix caching defaults to off."

    # with flag to turn it on.
    args = parser.parse_args(["--enable-prefix-caching"])
    engine_args = EngineArgs.from_cli_args(args=args)
    assert engine_args.enable_prefix_caching

    # with disable flag to turn it off.
    args = parser.parse_args(["--no-enable-prefix-caching"])
    engine_args = EngineArgs.from_cli_args(args=args)
    assert not engine_args.enable_prefix_caching


297
298
299
300
301
302
303
304
305
306
@pytest.mark.parametrize(
    ("arg"),
    [
        "image",  # Missing =
        "image=4,image=5",  # Conflicting values
        "image=video=4"  # Too many = in tokenized arg
    ])
def test_bad_nullable_kvs(arg):
    with pytest.raises(ArgumentTypeError):
        nullable_kvs(arg)
307
308


309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
# yapf: disable
@pytest.mark.parametrize(("arg", "expected", "option"), [
    (None, None, "mm-processor-kwargs"),
    ("{}", {}, "mm-processor-kwargs"),
    (
        '{"num_crops": 4}',
        {
            "num_crops": 4
        },
        "mm-processor-kwargs"
    ),
    (
        '{"foo": {"bar": "baz"}}',
        {
            "foo":
            {
                "bar": "baz"
            }
        },
        "mm-processor-kwargs"
    ),
    (
        '{"cast_logits_dtype":"bfloat16","sequence_parallel_norm":true,"sequence_parallel_norm_threshold":2048}',
        {
            "cast_logits_dtype": "bfloat16",
            "sequence_parallel_norm": True,
            "sequence_parallel_norm_threshold": 2048,
        },
        "override-neuron-config"
    ),
339
])
340
341
# yapf: enable
def test_composite_arg_parser(arg, expected, option):
342
343
344
345
    parser = EngineArgs.add_cli_args(FlexibleArgumentParser())
    if arg is None:
        args = parser.parse_args([])
    else:
346
347
        args = parser.parse_args([f"--{option}", arg])
    assert getattr(args, option.replace("-", "_")) == expected
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383


def test_human_readable_model_len():
    # `exit_on_error` disabled to test invalid values below
    parser = EngineArgs.add_cli_args(
        FlexibleArgumentParser(exit_on_error=False))

    args = parser.parse_args([])
    assert args.max_model_len is None

    args = parser.parse_args(["--max-model-len", "1024"])
    assert args.max_model_len == 1024

    # Lower
    args = parser.parse_args(["--max-model-len", "1m"])
    assert args.max_model_len == 1_000_000
    args = parser.parse_args(["--max-model-len", "10k"])
    assert args.max_model_len == 10_000

    # Capital
    args = parser.parse_args(["--max-model-len", "3K"])
    assert args.max_model_len == 1024 * 3
    args = parser.parse_args(["--max-model-len", "10M"])
    assert args.max_model_len == 2**20 * 10

    # Decimal values
    args = parser.parse_args(["--max-model-len", "10.2k"])
    assert args.max_model_len == 10200
    # ..truncated to the nearest int
    args = parser.parse_args(["--max-model-len", "10.212345k"])
    assert args.max_model_len == 10212

    # Invalid (do not allow decimals with binary multipliers)
    for invalid in ["1a", "pwd", "10.24", "1.23M"]:
        with pytest.raises(ArgumentError):
            args = parser.parse_args(["--max-model-len", invalid])