protocol.py 51.1 KB
Newer Older
1
2
# Adapted from
# https://github.com/lm-sys/FastChat/blob/168ccc29d3f7edc50823016105c024fe2282732a/fastchat/protocol/openai_api_protocol.py
3
import re
Zhuohan Li's avatar
Zhuohan Li committed
4
import time
5
from argparse import Namespace
6
from typing import Any, ClassVar, Dict, List, Literal, Optional, Set, Union
Zhuohan Li's avatar
Zhuohan Li committed
7

8
import torch
9
from pydantic import BaseModel, ConfigDict, Field, model_validator
10
from typing_extensions import Annotated
Zhuohan Li's avatar
Zhuohan Li committed
11

12
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
13
from vllm.logger import init_logger
14
from vllm.pooling_params import PoolingParams
15
16
from vllm.sampling_params import (BeamSearchParams, GuidedDecodingParams,
                                  RequestOutputKind, SamplingParams)
17
from vllm.sequence import Logprob
18
from vllm.utils import random_uuid, resolve_obj_by_qualname
19

20
21
logger = init_logger(__name__)

22
23
24
# torch is mocked during docs generation,
# so we have to provide the values as literals
_MOCK_LONG_INFO = Namespace(min=-9223372036854775808, max=9223372036854775807)
25
_LONG_INFO: Union["torch.iinfo", Namespace]
26
27
28
29
30
31
32
33
34
35
36
37
38
39

try:
    from sphinx.ext.autodoc.mock import _MockModule

    if isinstance(torch, _MockModule):
        _LONG_INFO = _MOCK_LONG_INFO
    else:
        _LONG_INFO = torch.iinfo(torch.long)
except ModuleNotFoundError:
    _LONG_INFO = torch.iinfo(torch.long)

assert _LONG_INFO.min == _MOCK_LONG_INFO.min
assert _LONG_INFO.max == _MOCK_LONG_INFO.max

Zhuohan Li's avatar
Zhuohan Li committed
40

41
class OpenAIBaseModel(BaseModel):
42
43
44
    # OpenAI API does allow extra fields
    model_config = ConfigDict(extra="allow")

45
46
47
    # Cache class field names
    field_names: ClassVar[Optional[Set[str]]] = None

48
49
50
    @model_validator(mode="before")
    @classmethod
    def __log_extra_fields__(cls, data):
51
52
53
54
55

        field_names = cls.field_names
        if field_names is None:
            if not isinstance(data, dict):
                return data
56
57
58
59
            # Get all class field names and their potential aliases
            field_names = set()
            for field_name, field in cls.model_fields.items():
                field_names.add(field_name)
60
61
62
63
64
65
66
67
68
69
                if alias := getattr(field, 'alias', None):
                    field_names.add(alias)
            cls.field_names = field_names

        # Compare against both field names and aliases
        if any(k not in field_names for k in data):
            logger.warning(
                "The following fields were present in the request "
                "but ignored: %s",
                data.keys() - field_names)
70
        return data
71
72
73


class ErrorResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
74
75
76
77
    object: str = "error"
    message: str
    type: str
    param: Optional[str] = None
78
    code: int
Zhuohan Li's avatar
Zhuohan Li committed
79
80


81
class ModelPermission(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
82
83
84
85
86
87
88
89
90
91
92
    id: str = Field(default_factory=lambda: f"modelperm-{random_uuid()}")
    object: str = "model_permission"
    created: int = Field(default_factory=lambda: int(time.time()))
    allow_create_engine: bool = False
    allow_sampling: bool = True
    allow_logprobs: bool = True
    allow_search_indices: bool = False
    allow_view: bool = True
    allow_fine_tuning: bool = False
    organization: str = "*"
    group: Optional[str] = None
93
    is_blocking: bool = False
Zhuohan Li's avatar
Zhuohan Li committed
94
95


96
class ModelCard(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
97
98
99
    id: str
    object: str = "model"
    created: int = Field(default_factory=lambda: int(time.time()))
Woosuk Kwon's avatar
Woosuk Kwon committed
100
    owned_by: str = "vllm"
Zhuohan Li's avatar
Zhuohan Li committed
101
102
    root: Optional[str] = None
    parent: Optional[str] = None
103
    max_model_len: Optional[int] = None
Zhuohan Li's avatar
Zhuohan Li committed
104
105
106
    permission: List[ModelPermission] = Field(default_factory=list)


107
class ModelList(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
108
109
110
111
    object: str = "list"
    data: List[ModelCard] = Field(default_factory=list)


112
113
114
115
class PromptTokenUsageInfo(OpenAIBaseModel):
    cached_tokens: Optional[int] = None


116
class UsageInfo(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
117
118
119
    prompt_tokens: int = 0
    total_tokens: int = 0
    completion_tokens: Optional[int] = 0
120
    prompt_tokens_details: Optional[PromptTokenUsageInfo] = None
Zhuohan Li's avatar
Zhuohan Li committed
121
122


123
124
125
126
127
class RequestResponseMetadata(BaseModel):
    request_id: str
    final_usage_info: Optional[UsageInfo] = None


128
129
130
131
132
133
134
135
136
class JsonSchemaResponseFormat(OpenAIBaseModel):
    name: str
    description: Optional[str] = None
    # schema is the field in openai but that causes conflicts with pydantic so
    # instead use json_schema with an alias
    json_schema: Optional[Dict[str, Any]] = Field(default=None, alias='schema')
    strict: Optional[bool] = None


137
class ResponseFormat(OpenAIBaseModel):
138
139
140
    # type must be "json_schema", "json_object" or "text"
    type: Literal["text", "json_object", "json_schema"]
    json_schema: Optional[JsonSchemaResponseFormat] = None
141
142


143
class StreamOptions(OpenAIBaseModel):
144
    include_usage: Optional[bool] = True
145
    continuous_usage_stats: Optional[bool] = False
146
147


148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
class FunctionDefinition(OpenAIBaseModel):
    name: str
    description: Optional[str] = None
    parameters: Optional[Dict[str, Any]] = None


class ChatCompletionToolsParam(OpenAIBaseModel):
    type: Literal["function"] = "function"
    function: FunctionDefinition


class ChatCompletionNamedFunction(OpenAIBaseModel):
    name: str


class ChatCompletionNamedToolChoiceParam(OpenAIBaseModel):
    function: ChatCompletionNamedFunction
    type: Literal["function"] = "function"


168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
class LogitsProcessorConstructor(BaseModel):
    qualname: str
    args: Optional[List[Any]] = None
    kwargs: Optional[Dict[str, Any]] = None


LogitsProcessors = List[Union[str, LogitsProcessorConstructor]]


def get_logits_processors(processors: Optional[LogitsProcessors],
                          pattern: Optional[str]) -> Optional[List[Any]]:
    if processors and pattern:
        logits_processors = []
        for processor in processors:
            qualname = processor if isinstance(processor,
                                               str) else processor.qualname
            if not re.match(pattern, qualname):
                raise ValueError(
                    f"Logits processor '{qualname}' is not allowed by this "
                    "server. See --logits-processor-pattern engine argument "
                    "for more information.")
            try:
                logits_processor = resolve_obj_by_qualname(qualname)
            except Exception as e:
                raise ValueError(
                    f"Logits processor '{qualname}' could not be resolved: {e}"
                ) from e
            if isinstance(processor, LogitsProcessorConstructor):
                logits_processor = logits_processor(*processor.args or [],
                                                    **processor.kwargs or {})
            logits_processors.append(logits_processor)
        return logits_processors
    elif processors:
        raise ValueError(
            "The `logits_processors` argument is not supported by this "
            "server. See --logits-processor-pattern engine argugment "
            "for more information.")
    return None


208
class ChatCompletionRequest(OpenAIBaseModel):
209
210
    # Ordered by official OpenAI API documentation
    # https://platform.openai.com/docs/api-reference/chat/create
211
    messages: List[ChatCompletionMessageParam]
212
213
214
215
    model: str
    frequency_penalty: Optional[float] = 0.0
    logit_bias: Optional[Dict[str, float]] = None
    logprobs: Optional[bool] = False
216
    top_logprobs: Optional[int] = 0
217
218
219
220
221
222
    # TODO(#9845): remove max_tokens when field is removed from OpenAI API
    max_tokens: Optional[int] = Field(
        default=None,
        deprecated=
        'max_tokens is deprecated in favor of the max_completion_tokens field')
    max_completion_tokens: Optional[int] = None
223
224
225
    n: Optional[int] = 1
    presence_penalty: Optional[float] = 0.0
    response_format: Optional[ResponseFormat] = None
226
    seed: Optional[int] = Field(None, ge=_LONG_INFO.min, le=_LONG_INFO.max)
227
    stop: Optional[Union[str, List[str]]] = Field(default_factory=list)
Zhuohan Li's avatar
Zhuohan Li committed
228
    stream: Optional[bool] = False
229
    stream_options: Optional[StreamOptions] = None
230
231
    temperature: Optional[float] = None
    top_p: Optional[float] = None
232
    tools: Optional[List[ChatCompletionToolsParam]] = None
233
    tool_choice: Optional[Union[Literal["none"], Literal["auto"],
234
                                ChatCompletionNamedToolChoiceParam]] = "none"
235
236
237

    # NOTE this will be ignored by VLLM -- the model determines the behavior
    parallel_tool_calls: Optional[bool] = False
Zhuohan Li's avatar
Zhuohan Li committed
238
    user: Optional[str] = None
239
240

    # doc: begin-chat-completion-sampling-params
241
    best_of: Optional[int] = None
242
    use_beam_search: bool = False
243
244
245
    top_k: Optional[int] = None
    min_p: Optional[float] = None
    repetition_penalty: Optional[float] = None
246
    length_penalty: float = 1.0
247
    stop_token_ids: Optional[List[int]] = Field(default_factory=list)
248
249
250
251
252
253
    include_stop_str_in_output: bool = False
    ignore_eos: bool = False
    min_tokens: int = 0
    skip_special_tokens: bool = True
    spaces_between_special_tokens: bool = True
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None
254
    prompt_logprobs: Optional[int] = None
255
256
257
    # doc: end-chat-completion-sampling-params

    # doc: begin-chat-completion-extra-params
258
    echo: bool = Field(
259
260
261
262
263
        default=False,
        description=(
            "If true, the new message will be prepended with the last message "
            "if they belong to the same role."),
    )
264
    add_generation_prompt: bool = Field(
265
266
267
268
269
270
        default=True,
        description=
        ("If true, the generation prompt will be added to the chat template. "
         "This is a parameter used by chat template in tokenizer config of the "
         "model."),
    )
271
272
273
274
275
276
277
278
279
    continue_final_message: bool = Field(
        default=False,
        description=
        ("If this is set, the chat will be formatted so that the final "
         "message in the chat is open-ended, without any EOS tokens. The "
         "model will continue this message rather than starting a new one. "
         "This allows you to \"prefill\" part of the model's response for it. "
         "Cannot be used at the same time as `add_generation_prompt`."),
    )
280
    add_special_tokens: bool = Field(
281
282
283
284
285
        default=False,
        description=(
            "If true, special tokens (e.g. BOS) will be added to the prompt "
            "on top of what is added by the chat template. "
            "For most models, the chat template takes care of adding the "
286
            "special tokens so this should be set to false (as is the "
287
288
            "default)."),
    )
289
290
291
292
293
294
295
296
297
298
299
300
301
    documents: Optional[List[Dict[str, str]]] = Field(
        default=None,
        description=
        ("A list of dicts representing documents that will be accessible to "
         "the model if it is performing RAG (retrieval-augmented generation)."
         " If the template does not support RAG, this argument will have no "
         "effect. We recommend that each document should be a dict containing "
         "\"title\" and \"text\" keys."),
    )
    chat_template: Optional[str] = Field(
        default=None,
        description=(
            "A Jinja template to use for this conversion. "
302
303
304
            "As of transformers v4.44, default chat template is no longer "
            "allowed, so you must provide a chat template if the tokenizer "
            "does not define one."),
305
306
307
308
309
310
    )
    chat_template_kwargs: Optional[Dict[str, Any]] = Field(
        default=None,
        description=("Additional kwargs to pass to the template renderer. "
                     "Will be accessible by the chat template."),
    )
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
    guided_json: Optional[Union[str, dict, BaseModel]] = Field(
        default=None,
        description=("If specified, the output will follow the JSON schema."),
    )
    guided_regex: Optional[str] = Field(
        default=None,
        description=(
            "If specified, the output will follow the regex pattern."),
    )
    guided_choice: Optional[List[str]] = Field(
        default=None,
        description=(
            "If specified, the output will be exactly one of the choices."),
    )
    guided_grammar: Optional[str] = Field(
        default=None,
        description=(
            "If specified, the output will follow the context free grammar."),
    )
330
331
332
333
334
335
    guided_decoding_backend: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default guided decoding backend "
            "of the server for this specific request. If set, must be either "
            "'outlines' / 'lm-format-enforcer'"))
336
337
338
339
340
    guided_whitespace_pattern: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default whitespace pattern "
            "for guided json decoding."))
341
342
343
344
345
346
    priority: int = Field(
        default=0,
        description=(
            "The priority of the request (lower means earlier handling; "
            "default: 0). Any priority other than 0 will raise an error "
            "if the served model does not use priority scheduling."))
347
348
349
350
351
352
    request_id: str = Field(
        default_factory=lambda: f"{random_uuid()}",
        description=(
            "The request_id related to this request. If the caller does "
            "not set it, a random_uuid will be generated. This id is used "
            "through out the inference process and return in response."))
353
354
355
356
357
358
359
360
361
362
363
    logits_processors: Optional[LogitsProcessors] = Field(
        default=None,
        description=(
            "A list of either qualified names of logits processors, or "
            "constructor objects, to apply when sampling. A constructor is "
            "a JSON object with a required 'qualname' field specifying the "
            "qualified name of the processor class/factory, and optional "
            "'args' and 'kwargs' fields containing positional and keyword "
            "arguments. For example: {'qualname': "
            "'my_module.MyLogitsProcessor', 'args': [1, 2], 'kwargs': "
            "{'param': 'value'}}."))
364
365

    # doc: end-chat-completion-extra-params
Zhuohan Li's avatar
Zhuohan Li committed
366

367
368
369
370
371
372
373
374
375
376
377
378
379
380
    # Default sampling parameters for chat completion requests
    _DEFAULT_SAMPLING_PARAMS: dict = {
        "repetition_penalty": 1.0,
        "temperature": 1.0,
        "top_p": 1.0,
        "top_k": -1,
        "min_p": 0.0,
    }

    def to_beam_search_params(
            self,
            default_max_tokens: int,
            default_sampling_params: Optional[dict] = None
    ) -> BeamSearchParams:
381
382
        # TODO(#9845): remove max_tokens when field is removed from OpenAI API
        max_tokens = self.max_completion_tokens or self.max_tokens
383

384
385
        if default_sampling_params is None:
            default_sampling_params = {}
386
        n = self.n if self.n is not None else 1
387

388
389
390
391
392
393
        # Use minimum of context window, user request & server limit.
        max_tokens = min(
            val for val in (default_max_tokens, max_tokens,
                            default_sampling_params.get("max_tokens", None))
            if val is not None)

394
395
396
        if (temperature := self.temperature) is None:
            temperature = default_sampling_params.get(
                "temperature", self._DEFAULT_SAMPLING_PARAMS["temperature"])
397
398
399
400
401
402

        return BeamSearchParams(
            beam_width=n,
            max_tokens=max_tokens,
            ignore_eos=self.ignore_eos,
            temperature=temperature,
403
            length_penalty=self.length_penalty,
404
            include_stop_str_in_output=self.include_stop_str_in_output)
405

406
    def to_sampling_params(
407
408
409
410
            self,
            default_max_tokens: int,
            logits_processor_pattern: Optional[str],
            default_sampling_params: Optional[dict] = None) -> SamplingParams:
411
412
        # TODO(#9845): remove max_tokens when field is removed from OpenAI API
        max_tokens = self.max_completion_tokens or self.max_tokens
413

414
415
        if default_sampling_params is None:
            default_sampling_params = {}
416
417
418
419
420
421
422

        # Use minimum of context window, user request & server limit.
        max_tokens = min(
            val for val in (default_max_tokens, max_tokens,
                            default_sampling_params.get("max_tokens", None))
            if val is not None)

423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
        # Default parameters
        if (repetition_penalty := self.repetition_penalty) is None:
            repetition_penalty = default_sampling_params.get(
                "repetition_penalty",
                self._DEFAULT_SAMPLING_PARAMS["repetition_penalty"],
            )
        if (temperature := self.temperature) is None:
            temperature = default_sampling_params.get(
                "temperature", self._DEFAULT_SAMPLING_PARAMS["temperature"])
        if (top_p := self.top_p) is None:
            top_p = default_sampling_params.get(
                "top_p", self._DEFAULT_SAMPLING_PARAMS["top_p"])
        if (top_k := self.top_k) is None:
            top_k = default_sampling_params.get(
                "top_k", self._DEFAULT_SAMPLING_PARAMS["top_k"])
        if (min_p := self.min_p) is None:
            min_p = default_sampling_params.get(
                "min_p", self._DEFAULT_SAMPLING_PARAMS["min_p"])

442
443
444
445
        prompt_logprobs = self.prompt_logprobs
        if prompt_logprobs is None and self.echo:
            prompt_logprobs = self.top_logprobs

446
        guided_json_object = None
447
448
449
450
451
452
453
454
        if self.response_format is not None:
            if self.response_format.type == "json_object":
                guided_json_object = True
            elif self.response_format.type == "json_schema":
                json_schema = self.response_format.json_schema
                assert json_schema is not None
                self.guided_json = json_schema.json_schema
                if self.guided_decoding_backend is None:
455
                    self.guided_decoding_backend = "xgrammar"
456
457
458
459
460
461
462
463
464

        guided_decoding = GuidedDecodingParams.from_optional(
            json=self._get_guided_json_from_tool() or self.guided_json,
            regex=self.guided_regex,
            choice=self.guided_choice,
            grammar=self.guided_grammar,
            json_object=guided_json_object,
            backend=self.guided_decoding_backend,
            whitespace_pattern=self.guided_whitespace_pattern)
465

466
        return SamplingParams.from_optional(
467
            n=self.n,
468
            best_of=self.best_of,
469
470
            presence_penalty=self.presence_penalty,
            frequency_penalty=self.frequency_penalty,
471
472
473
474
475
            repetition_penalty=repetition_penalty,
            temperature=temperature,
            top_p=top_p,
            top_k=top_k,
            min_p=min_p,
Nick Hill's avatar
Nick Hill committed
476
            seed=self.seed,
477
478
            stop=self.stop,
            stop_token_ids=self.stop_token_ids,
479
            logprobs=self.top_logprobs if self.logprobs else None,
480
            prompt_logprobs=prompt_logprobs,
481
            ignore_eos=self.ignore_eos,
482
            max_tokens=max_tokens,
483
            min_tokens=self.min_tokens,
484
485
            skip_special_tokens=self.skip_special_tokens,
            spaces_between_special_tokens=self.spaces_between_special_tokens,
486
487
            logits_processors=get_logits_processors(self.logits_processors,
                                                    logits_processor_pattern),
488
            include_stop_str_in_output=self.include_stop_str_in_output,
489
            truncate_prompt_tokens=self.truncate_prompt_tokens,
490
491
            output_kind=RequestOutputKind.DELTA if self.stream \
                else RequestOutputKind.FINAL_ONLY,
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
            guided_decoding=guided_decoding,
            logit_bias=self.logit_bias)

    def _get_guided_json_from_tool(
            self) -> Optional[Union[str, dict, BaseModel]]:
        # user has chosen to not use any tool
        if self.tool_choice == "none" or self.tools is None:
            return None

        # user has chosen to use a named tool
        if type(self.tool_choice) is ChatCompletionNamedToolChoiceParam:
            tool_name = self.tool_choice.function.name
            tools = {tool.function.name: tool.function for tool in self.tools}
            if tool_name not in tools:
                raise ValueError(
                    f"Tool '{tool_name}' has not been passed in `tools`.")
            tool = tools[tool_name]
            return tool.parameters

        return None
512

513
    @model_validator(mode="before")
514
    @classmethod
515
516
    def validate_stream_options(cls, data):
        if data.get("stream_options") and not data.get("stream"):
517
            raise ValueError(
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
                "Stream options can only be defined when `stream=True`.")

        return data

    @model_validator(mode="before")
    @classmethod
    def check_logprobs(cls, data):
        if (prompt_logprobs := data.get("prompt_logprobs")) is not None:
            if data.get("stream") and prompt_logprobs > 0:
                raise ValueError(
                    "`prompt_logprobs` are not available when `stream=True`.")

            if prompt_logprobs < 0:
                raise ValueError("`prompt_logprobs` must be a positive value.")

        if (top_logprobs := data.get("top_logprobs")) is not None:
            if top_logprobs < 0:
                raise ValueError("`top_logprobs` must be a positive value.")

            if not data.get("logprobs"):
                raise ValueError(
                    "when using `top_logprobs`, `logprobs` must be set to true."
                )

        return data
543

544
545
546
    @model_validator(mode="before")
    @classmethod
    def check_guided_decoding_count(cls, data):
547
548
549
        if isinstance(data, ValueError):
            raise data

550
551
552
553
554
        guide_count = sum([
            "guided_json" in data and data["guided_json"] is not None,
            "guided_regex" in data and data["guided_regex"] is not None,
            "guided_choice" in data and data["guided_choice"] is not None
        ])
555
        # you can only use one kind of guided decoding
556
557
558
559
        if guide_count > 1:
            raise ValueError(
                "You can only use one kind of guided decoding "
                "('guided_json', 'guided_regex' or 'guided_choice').")
560
        # you can only either use guided decoding or tools, not both
561
562
        if guide_count > 1 and data.get("tool_choice",
                                        "none") not in ("none", "auto"):
563
564
565
566
567
568
            raise ValueError(
                "You can only either use guided decoding or tools, not both.")
        return data

    @model_validator(mode="before")
    @classmethod
569
570
571
572
    def check_tool_usage(cls, data):

        # if "tool_choice" is not specified but tools are provided,
        # default to "auto" tool_choice
573
        if "tool_choice" not in data and data.get("tools"):
574
575
            data["tool_choice"] = "auto"

576
577
578
579
580
581
        # if "tool_choice" is "none" -- ignore tools if present
        if "tool_choice" in data and data["tool_choice"] == "none":
            # ensure that no tools are present
            data.pop("tools", None)
            return data

582
583
584
585
        # if "tool_choice" is specified -- validation
        if "tool_choice" in data:

            # ensure that if "tool choice" is specified, tools are present
586
587
588
            if "tools" not in data or data["tools"] is None:
                raise ValueError(
                    "When using `tool_choice`, `tools` must be set.")
589
590
591
592
593
594

            # make sure that tool choice is either a named tool
            # OR that it's set to "auto"
            if data["tool_choice"] != "auto" and not isinstance(
                    data["tool_choice"], dict):
                raise ValueError(
595
596
                    "`tool_choice` must either be a named tool, \"auto\", "
                    "or \"none\".")
597
598
599
600
601

            # ensure that if "tool_choice" is specified as an object,
            # it matches a valid tool
            if isinstance(data["tool_choice"], dict):
                valid_tool = False
602
                specified_function = data["tool_choice"].get("function")
603
604
                if not specified_function:
                    raise ValueError(
605
606
                        "Expected field `function` in `tool_choice`."
                        " Correct usage: `{\"type\": \"function\","
607
                        " \"function\": {\"name\": \"my_function\"}}`")
608
                specified_function_name = specified_function.get("name")
609
610
                if not specified_function_name:
                    raise ValueError(
611
612
                        "Expected field `name` in `function` in `tool_choice`."
                        "Correct usage: `{\"type\": \"function\", "
613
614
615
616
617
618
619
620
621
                        "\"function\": {\"name\": \"my_function\"}}`")
                for tool in data["tools"]:
                    if tool["function"]["name"] == specified_function_name:
                        valid_tool = True
                        break
                if not valid_tool:
                    raise ValueError(
                        "The tool specified in `tool_choice` does not match any"
                        " of the specified `tools`")
622
623
        return data

624
625
626
627
628
629
630
631
632
    @model_validator(mode="before")
    @classmethod
    def check_generation_prompt(cls, data):
        if data.get("continue_final_message") and data.get(
                "add_generation_prompt"):
            raise ValueError("Cannot set both `continue_final_message` and "
                             "`add_generation_prompt` to True.")
        return data

Zhuohan Li's avatar
Zhuohan Li committed
633

634
class CompletionRequest(OpenAIBaseModel):
635
636
    # Ordered by official OpenAI API documentation
    # https://platform.openai.com/docs/api-reference/completions/create
Zhuohan Li's avatar
Zhuohan Li committed
637
    model: str
638
    prompt: Union[List[int], List[List[int]], str, List[str]]
639
    best_of: Optional[int] = None
Zhuohan Li's avatar
Zhuohan Li committed
640
641
642
    echo: Optional[bool] = False
    frequency_penalty: Optional[float] = 0.0
    logit_bias: Optional[Dict[str, float]] = None
643
644
    logprobs: Optional[int] = None
    max_tokens: Optional[int] = 16
645
    n: int = 1
646
    presence_penalty: Optional[float] = 0.0
647
    seed: Optional[int] = Field(None, ge=_LONG_INFO.min, le=_LONG_INFO.max)
648
649
    stop: Optional[Union[str, List[str]]] = Field(default_factory=list)
    stream: Optional[bool] = False
650
    stream_options: Optional[StreamOptions] = None
651
    suffix: Optional[str] = None
652
653
    temperature: Optional[float] = None
    top_p: Optional[float] = None
Zhuohan Li's avatar
Zhuohan Li committed
654
    user: Optional[str] = None
655
656

    # doc: begin-completion-sampling-params
657
    use_beam_search: bool = False
658
659
660
    top_k: Optional[int] = None
    min_p: Optional[float] = None
    repetition_penalty: Optional[float] = None
661
    length_penalty: float = 1.0
662
    stop_token_ids: Optional[List[int]] = Field(default_factory=list)
663
664
665
666
667
    include_stop_str_in_output: bool = False
    ignore_eos: bool = False
    min_tokens: int = 0
    skip_special_tokens: bool = True
    spaces_between_special_tokens: bool = True
668
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None
669
    allowed_token_ids: Optional[List[int]] = None
670
    prompt_logprobs: Optional[int] = None
671
672
673
    # doc: end-completion-sampling-params

    # doc: begin-completion-extra-params
674
675
    add_special_tokens: bool = Field(
        default=True,
676
        description=(
677
678
            "If true (the default), special tokens (e.g. BOS) will be added to "
            "the prompt."),
679
680
681
682
683
    )
    response_format: Optional[ResponseFormat] = Field(
        default=None,
        description=
        ("Similar to chat completion, this parameter specifies the format of "
684
685
         "output. Only {'type': 'json_object'}, {'type': 'json_schema'} or "
         "{'type': 'text' } is supported."),
686
687
688
    )
    guided_json: Optional[Union[str, dict, BaseModel]] = Field(
        default=None,
689
        description="If specified, the output will follow the JSON schema.",
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
    )
    guided_regex: Optional[str] = Field(
        default=None,
        description=(
            "If specified, the output will follow the regex pattern."),
    )
    guided_choice: Optional[List[str]] = Field(
        default=None,
        description=(
            "If specified, the output will be exactly one of the choices."),
    )
    guided_grammar: Optional[str] = Field(
        default=None,
        description=(
            "If specified, the output will follow the context free grammar."),
    )
706
707
708
709
710
711
    guided_decoding_backend: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default guided decoding backend "
            "of the server for this specific request. If set, must be one of "
            "'outlines' / 'lm-format-enforcer'"))
712
713
714
715
716
    guided_whitespace_pattern: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default whitespace pattern "
            "for guided json decoding."))
717
718
719
720
721
722
    priority: int = Field(
        default=0,
        description=(
            "The priority of the request (lower means earlier handling; "
            "default: 0). Any priority other than 0 will raise an error "
            "if the served model does not use priority scheduling."))
723
724
725
726
727
728
729
730
731
732
733
    logits_processors: Optional[LogitsProcessors] = Field(
        default=None,
        description=(
            "A list of either qualified names of logits processors, or "
            "constructor objects, to apply when sampling. A constructor is "
            "a JSON object with a required 'qualname' field specifying the "
            "qualified name of the processor class/factory, and optional "
            "'args' and 'kwargs' fields containing positional and keyword "
            "arguments. For example: {'qualname': "
            "'my_module.MyLogitsProcessor', 'args': [1, 2], 'kwargs': "
            "{'param': 'value'}}."))
734
735

    # doc: end-completion-extra-params
Zhuohan Li's avatar
Zhuohan Li committed
736

737
738
739
740
741
742
743
744
745
746
747
748
749
750
    # Default sampling parameters for completion requests
    _DEFAULT_SAMPLING_PARAMS: dict = {
        "repetition_penalty": 1.0,
        "temperature": 1.0,
        "top_p": 1.0,
        "top_k": -1,
        "min_p": 0.0,
    }

    def to_beam_search_params(
            self,
            default_max_tokens: int,
            default_sampling_params: Optional[dict] = None
    ) -> BeamSearchParams:
751
752
        max_tokens = self.max_tokens

753
754
        if default_sampling_params is None:
            default_sampling_params = {}
755
        n = self.n if self.n is not None else 1
756

757
758
759
760
761
762
        # Use minimum of context window, user request & server limit.
        max_tokens = min(
            val for val in (default_max_tokens, max_tokens,
                            default_sampling_params.get("max_tokens", None))
            if val is not None)

763
764
        if (temperature := self.temperature) is None:
            temperature = default_sampling_params.get("temperature", 1.0)
765
766
767
768
769
770

        return BeamSearchParams(
            beam_width=n,
            max_tokens=max_tokens,
            ignore_eos=self.ignore_eos,
            temperature=temperature,
771
            length_penalty=self.length_penalty,
772
            include_stop_str_in_output=self.include_stop_str_in_output)
773

774
    def to_sampling_params(
775
776
777
778
            self,
            default_max_tokens: int,
            logits_processor_pattern: Optional[str],
            default_sampling_params: Optional[dict] = None) -> SamplingParams:
779
780
        max_tokens = self.max_tokens

781
782
        if default_sampling_params is None:
            default_sampling_params = {}
783
784
785
786
787
788
789

        # Use minimum of context window, user request & server limit.
        max_tokens = min(
            val for val in (default_max_tokens, max_tokens,
                            default_sampling_params.get("max_tokens", None))
            if val is not None)

790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
        # Default parameters
        if (repetition_penalty := self.repetition_penalty) is None:
            repetition_penalty = default_sampling_params.get(
                "repetition_penalty",
                self._DEFAULT_SAMPLING_PARAMS["repetition_penalty"],
            )
        if (temperature := self.temperature) is None:
            temperature = default_sampling_params.get(
                "temperature", self._DEFAULT_SAMPLING_PARAMS["temperature"])
        if (top_p := self.top_p) is None:
            top_p = default_sampling_params.get(
                "top_p", self._DEFAULT_SAMPLING_PARAMS["top_p"])
        if (top_k := self.top_k) is None:
            top_k = default_sampling_params.get(
                "top_k", self._DEFAULT_SAMPLING_PARAMS["top_k"])
        if (min_p := self.min_p) is None:
            min_p = default_sampling_params.get(
                "min_p", self._DEFAULT_SAMPLING_PARAMS["min_p"])

809
810
811
812
        prompt_logprobs = self.prompt_logprobs
        if prompt_logprobs is None and self.echo:
            prompt_logprobs = self.logprobs

813
814
        echo_without_generation = self.echo and self.max_tokens == 0

815
816
817
818
819
820
821
822
823
824
825
826
827
        guided_json_object = None
        if (self.response_format is not None
                and self.response_format.type == "json_object"):
            guided_json_object = True

        guided_decoding = GuidedDecodingParams.from_optional(
            json=self.guided_json,
            regex=self.guided_regex,
            choice=self.guided_choice,
            grammar=self.guided_grammar,
            json_object=guided_json_object,
            backend=self.guided_decoding_backend,
            whitespace_pattern=self.guided_whitespace_pattern)
828

829
        return SamplingParams.from_optional(
830
831
832
833
            n=self.n,
            best_of=self.best_of,
            presence_penalty=self.presence_penalty,
            frequency_penalty=self.frequency_penalty,
834
835
836
837
838
            repetition_penalty=repetition_penalty,
            temperature=temperature,
            top_p=top_p,
            top_k=top_k,
            min_p=min_p,
Nick Hill's avatar
Nick Hill committed
839
            seed=self.seed,
840
841
            stop=self.stop,
            stop_token_ids=self.stop_token_ids,
842
            logprobs=self.logprobs,
843
            ignore_eos=self.ignore_eos,
844
            max_tokens=max_tokens if not echo_without_generation else 1,
845
            min_tokens=self.min_tokens,
846
            prompt_logprobs=prompt_logprobs,
847
            skip_special_tokens=self.skip_special_tokens,
848
            spaces_between_special_tokens=self.spaces_between_special_tokens,
849
            include_stop_str_in_output=self.include_stop_str_in_output,
850
851
            logits_processors=get_logits_processors(self.logits_processors,
                                                    logits_processor_pattern),
852
            truncate_prompt_tokens=self.truncate_prompt_tokens,
853
854
            output_kind=RequestOutputKind.DELTA if self.stream \
                else RequestOutputKind.FINAL_ONLY,
855
856
857
            guided_decoding=guided_decoding,
            logit_bias=self.logit_bias,
            allowed_token_ids=self.allowed_token_ids)
858

859
860
861
862
863
864
865
866
867
868
869
870
871
872
    @model_validator(mode="before")
    @classmethod
    def check_guided_decoding_count(cls, data):
        guide_count = sum([
            "guided_json" in data and data["guided_json"] is not None,
            "guided_regex" in data and data["guided_regex"] is not None,
            "guided_choice" in data and data["guided_choice"] is not None
        ])
        if guide_count > 1:
            raise ValueError(
                "You can only use one kind of guided decoding "
                "('guided_json', 'guided_regex' or 'guided_choice').")
        return data

873
874
875
    @model_validator(mode="before")
    @classmethod
    def check_logprobs(cls, data):
876
877
878
879
880
881
882
883
884
885
886
        if (prompt_logprobs := data.get("prompt_logprobs")) is not None:
            if data.get("stream") and prompt_logprobs > 0:
                raise ValueError(
                    "`prompt_logprobs` are not available when `stream=True`.")

            if prompt_logprobs < 0:
                raise ValueError("`prompt_logprobs` must be a positive value.")

        if (logprobs := data.get("logprobs")) is not None and logprobs < 0:
            raise ValueError("`logprobs` must be a positive value.")

887
888
        return data

889
890
891
892
893
    @model_validator(mode="before")
    @classmethod
    def validate_stream_options(cls, data):
        if data.get("stream_options") and not data.get("stream"):
            raise ValueError(
894
895
                "Stream options can only be defined when `stream=True`.")

896
897
        return data

Zhuohan Li's avatar
Zhuohan Li committed
898

899
class EmbeddingCompletionRequest(OpenAIBaseModel):
900
901
902
903
    # Ordered by official OpenAI API documentation
    # https://platform.openai.com/docs/api-reference/embeddings
    model: str
    input: Union[List[int], List[List[int]], str, List[str]]
904
    encoding_format: Literal["float", "base64"] = "float"
905
906
    dimensions: Optional[int] = None
    user: Optional[str] = None
907
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None
908
909
910
911
912

    # doc: begin-embedding-pooling-params
    additional_data: Optional[Any] = None
    # doc: end-embedding-pooling-params

913
    # doc: begin-embedding-extra-params
914
915
916
917
918
919
    add_special_tokens: bool = Field(
        default=True,
        description=(
            "If true (the default), special tokens (e.g. BOS) will be added to "
            "the prompt."),
    )
920
921
922
923
924
925
926
927
928
    priority: int = Field(
        default=0,
        description=(
            "The priority of the request (lower means earlier handling; "
            "default: 0). Any priority other than 0 will raise an error "
            "if the served model does not use priority scheduling."))

    # doc: end-embedding-extra-params

929
930
931
932
    def to_pooling_params(self):
        return PoolingParams(additional_data=self.additional_data)


933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
class EmbeddingChatRequest(OpenAIBaseModel):
    model: str
    messages: List[ChatCompletionMessageParam]

    encoding_format: Literal["float", "base64"] = "float"
    dimensions: Optional[int] = None
    user: Optional[str] = None
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None

    # doc: begin-chat-embedding-pooling-params
    additional_data: Optional[Any] = None
    # doc: end-chat-embedding-pooling-params

    # doc: begin-chat-embedding-extra-params
    add_special_tokens: bool = Field(
        default=False,
        description=(
            "If true, special tokens (e.g. BOS) will be added to the prompt "
            "on top of what is added by the chat template. "
            "For most models, the chat template takes care of adding the "
            "special tokens so this should be set to false (as is the "
            "default)."),
    )
    chat_template: Optional[str] = Field(
        default=None,
        description=(
            "A Jinja template to use for this conversion. "
            "As of transformers v4.44, default chat template is no longer "
            "allowed, so you must provide a chat template if the tokenizer "
            "does not define one."),
    )
    chat_template_kwargs: Optional[Dict[str, Any]] = Field(
        default=None,
        description=("Additional kwargs to pass to the template renderer. "
                     "Will be accessible by the chat template."),
    )
    priority: int = Field(
        default=0,
        description=(
            "The priority of the request (lower means earlier handling; "
            "default: 0). Any priority other than 0 will raise an error "
            "if the served model does not use priority scheduling."))
    # doc: end-chat-embedding-extra-params

    @model_validator(mode="before")
    @classmethod
    def check_generation_prompt(cls, data):
        if data.get("continue_final_message") and data.get(
                "add_generation_prompt"):
            raise ValueError("Cannot set both `continue_final_message` and "
                             "`add_generation_prompt` to True.")
        return data

    def to_pooling_params(self):
        return PoolingParams(additional_data=self.additional_data)


EmbeddingRequest = Union[EmbeddingCompletionRequest, EmbeddingChatRequest]

992
993
994
995
PoolingCompletionRequest = EmbeddingCompletionRequest
PoolingChatRequest = EmbeddingChatRequest
PoolingRequest = Union[PoolingCompletionRequest, PoolingChatRequest]

996

997
998
999
1000
1001
1002
class ScoreRequest(OpenAIBaseModel):
    model: str
    text_1: Union[List[str], str]
    text_2: Union[List[str], str]
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None

1003
    # doc: begin-score-pooling-params
1004
    additional_data: Optional[Any] = None
1005
    # doc: end-score-pooling-params
1006

1007
    # doc: begin-score-extra-params
1008
1009
1010
1011
1012
1013
1014
    priority: int = Field(
        default=0,
        description=(
            "The priority of the request (lower means earlier handling; "
            "default: 0). Any priority other than 0 will raise an error "
            "if the served model does not use priority scheduling."))

1015
1016
    # doc: end-score-extra-params

1017
1018
1019
1020
    def to_pooling_params(self):
        return PoolingParams(additional_data=self.additional_data)


1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
1059
1060
1061
1062
1063
1064
1065
1066
class RerankRequest(OpenAIBaseModel):
    model: str
    query: str
    documents: List[str]
    top_n: int = Field(default_factory=lambda: 0)
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None

    # doc: begin-rerank-pooling-params
    additional_data: Optional[Any] = None
    # doc: end-rerank-pooling-params

    # doc: begin-rerank-extra-params
    priority: int = Field(
        default=0,
        description=(
            "The priority of the request (lower means earlier handling; "
            "default: 0). Any priority other than 0 will raise an error "
            "if the served model does not use priority scheduling."))

    # doc: end-rerank-extra-params

    def to_pooling_params(self):
        return PoolingParams(additional_data=self.additional_data)


class RerankDocument(BaseModel):
    text: str


class RerankResult(BaseModel):
    index: int
    document: RerankDocument
    relevance_score: float


class RerankUsage(BaseModel):
    total_tokens: int


class RerankResponse(OpenAIBaseModel):
    id: str
    model: str
    usage: RerankUsage
    results: List[RerankResult]


1067
class CompletionLogProbs(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1068
1069
1070
    text_offset: List[int] = Field(default_factory=list)
    token_logprobs: List[Optional[float]] = Field(default_factory=list)
    tokens: List[str] = Field(default_factory=list)
1071
1072
    top_logprobs: List[Optional[Dict[str,
                                     float]]] = Field(default_factory=list)
Zhuohan Li's avatar
Zhuohan Li committed
1073
1074


1075
class CompletionResponseChoice(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1076
1077
    index: int
    text: str
1078
    logprobs: Optional[CompletionLogProbs] = None
1079
1080
    finish_reason: Optional[str] = None
    stop_reason: Optional[Union[int, str]] = Field(
1081
1082
1083
1084
1085
1086
        default=None,
        description=(
            "The stop string or token id that caused the completion "
            "to stop, None if the completion finished for some other reason "
            "including encountering the EOS token"),
    )
1087
    prompt_logprobs: Optional[List[Optional[Dict[int, Logprob]]]] = None
Zhuohan Li's avatar
Zhuohan Li committed
1088
1089


1090
class CompletionResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1091
1092
1093
1094
1095
1096
1097
1098
    id: str = Field(default_factory=lambda: f"cmpl-{random_uuid()}")
    object: str = "text_completion"
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[CompletionResponseChoice]
    usage: UsageInfo


1099
class CompletionResponseStreamChoice(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1100
1101
    index: int
    text: str
1102
    logprobs: Optional[CompletionLogProbs] = None
1103
1104
    finish_reason: Optional[str] = None
    stop_reason: Optional[Union[int, str]] = Field(
1105
1106
1107
1108
1109
1110
        default=None,
        description=(
            "The stop string or token id that caused the completion "
            "to stop, None if the completion finished for some other reason "
            "including encountering the EOS token"),
    )
Zhuohan Li's avatar
Zhuohan Li committed
1111
1112


1113
class CompletionStreamResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1114
1115
1116
1117
1118
    id: str = Field(default_factory=lambda: f"cmpl-{random_uuid()}")
    object: str = "text_completion"
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[CompletionResponseStreamChoice]
1119
    usage: Optional[UsageInfo] = Field(default=None)
1120
1121


1122
class EmbeddingResponseData(OpenAIBaseModel):
1123
1124
    index: int
    object: str = "embedding"
1125
    embedding: Union[List[float], str]
1126
1127


1128
class EmbeddingResponse(OpenAIBaseModel):
1129
    id: str = Field(default_factory=lambda: f"embd-{random_uuid()}")
1130
1131
1132
1133
1134
1135
1136
    object: str = "list"
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    data: List[EmbeddingResponseData]
    usage: UsageInfo


1137
1138
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
class PoolingResponseData(OpenAIBaseModel):
    index: int
    object: str = "pooling"
    data: Union[List[List[float]], List[float], str]


class PoolingResponse(OpenAIBaseModel):
    id: str = Field(default_factory=lambda: f"pool-{random_uuid()}")
    object: str = "list"
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    data: List[PoolingResponseData]
    usage: UsageInfo


1152
1153
1154
class ScoreResponseData(OpenAIBaseModel):
    index: int
    object: str = "score"
1155
    score: float
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166


class ScoreResponse(OpenAIBaseModel):
    id: str = Field(default_factory=lambda: f"embd-{random_uuid()}")
    object: str = "list"
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    data: List[ScoreResponseData]
    usage: UsageInfo


1167
1168
1169
1170
1171
1172
1173
1174
1175
1176
1177
class FunctionCall(OpenAIBaseModel):
    name: str
    arguments: str


class ToolCall(OpenAIBaseModel):
    id: str = Field(default_factory=lambda: f"chatcmpl-tool-{random_uuid()}")
    type: Literal["function"] = "function"
    function: FunctionCall


1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
class DeltaFunctionCall(BaseModel):
    name: Optional[str] = None
    arguments: Optional[str] = None


# a tool call delta where everything is optional
class DeltaToolCall(OpenAIBaseModel):
    id: str = Field(default_factory=lambda: f"chatcmpl-tool-{random_uuid()}")
    type: Literal["function"] = "function"
    index: int
    function: Optional[DeltaFunctionCall] = None


class ExtractedToolCallInformation(BaseModel):
    # indicate if tools were called
    tools_called: bool

    # extracted tool calls
    tool_calls: List[ToolCall]

    # content - per OpenAI spec, content AND tool calls can be returned rarely
    # But some models will do this intentionally
    content: Optional[str] = None


1203
class ChatMessage(OpenAIBaseModel):
1204
    role: str
1205
    reasoning_content: Optional[str] = None
1206
    content: Optional[str] = None
1207
    tool_calls: List[ToolCall] = Field(default_factory=list)
1208
1209


1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
class ChatCompletionLogProb(OpenAIBaseModel):
    token: str
    logprob: float = -9999.0
    bytes: Optional[List[int]] = None


class ChatCompletionLogProbsContent(ChatCompletionLogProb):
    top_logprobs: List[ChatCompletionLogProb] = Field(default_factory=list)


class ChatCompletionLogProbs(OpenAIBaseModel):
    content: Optional[List[ChatCompletionLogProbsContent]] = None


1224
class ChatCompletionResponseChoice(OpenAIBaseModel):
1225
1226
    index: int
    message: ChatMessage
1227
    logprobs: Optional[ChatCompletionLogProbs] = None
1228
1229
1230
    # per OpenAI spec this is the default
    finish_reason: Optional[str] = "stop"
    # not part of the OpenAI spec but included in vLLM for legacy reasons
1231
    stop_reason: Optional[Union[int, str]] = None
1232
1233


1234
class ChatCompletionResponse(OpenAIBaseModel):
1235
    id: str = Field(default_factory=lambda: f"chatcmpl-{random_uuid()}")
1236
    object: Literal["chat.completion"] = "chat.completion"
1237
1238
1239
1240
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[ChatCompletionResponseChoice]
    usage: UsageInfo
1241
    prompt_logprobs: Optional[List[Optional[Dict[int, Logprob]]]] = None
1242
1243


1244
class DeltaMessage(OpenAIBaseModel):
1245
1246
    role: Optional[str] = None
    content: Optional[str] = None
1247
    reasoning_content: Optional[str] = None
1248
    tool_calls: List[DeltaToolCall] = Field(default_factory=list)
1249
1250


1251
class ChatCompletionResponseStreamChoice(OpenAIBaseModel):
1252
1253
    index: int
    delta: DeltaMessage
1254
    logprobs: Optional[ChatCompletionLogProbs] = None
1255
    finish_reason: Optional[str] = None
1256
    stop_reason: Optional[Union[int, str]] = None
1257
1258


1259
class ChatCompletionStreamResponse(OpenAIBaseModel):
1260
    id: str = Field(default_factory=lambda: f"chatcmpl-{random_uuid()}")
1261
    object: Literal["chat.completion.chunk"] = "chat.completion.chunk"
1262
1263
1264
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[ChatCompletionResponseStreamChoice]
1265
    usage: Optional[UsageInfo] = Field(default=None)
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286


class BatchRequestInput(OpenAIBaseModel):
    """
    The per-line object of the batch input file.

    NOTE: Currently only the `/v1/chat/completions` endpoint is supported.
    """

    # A developer-provided per-request id that will be used to match outputs to
    # inputs. Must be unique for each request in a batch.
    custom_id: str

    # The HTTP method to be used for the request. Currently only POST is
    # supported.
    method: str

    # The OpenAI API relative URL to be used for the request. Currently
    # /v1/chat/completions is supported.
    url: str

1287
    # The parameters of the request.
1288
    body: Union[ChatCompletionRequest, EmbeddingRequest, ScoreRequest]
1289
1290


1291
1292
1293
1294
1295
1296
1297
1298
class BatchResponseData(OpenAIBaseModel):
    # HTTP status code of the response.
    status_code: int = 200

    # An unique identifier for the API request.
    request_id: str

    # The body of the response.
1299
1300
    body: Optional[Union[ChatCompletionResponse, EmbeddingResponse,
                         ScoreResponse]] = None
1301
1302


1303
1304
1305
1306
1307
1308
1309
1310
1311
1312
1313
class BatchRequestOutput(OpenAIBaseModel):
    """
    The per-line object of the batch output and error files
    """

    id: str

    # A developer-provided per-request id that will be used to match outputs to
    # inputs.
    custom_id: str

1314
    response: Optional[BatchResponseData]
1315
1316
1317
1318

    # For requests that failed with a non-HTTP error, this will contain more
    # information on the cause of the failure.
    error: Optional[Any]
1319
1320


1321
1322
1323
1324
class TokenizeCompletionRequest(OpenAIBaseModel):
    model: str
    prompt: str

1325
1326
1327
1328
1329
1330
    add_special_tokens: bool = Field(
        default=True,
        description=(
            "If true (the default), special tokens (e.g. BOS) will be added to "
            "the prompt."),
    )
1331
1332
1333
1334
1335
1336


class TokenizeChatRequest(OpenAIBaseModel):
    model: str
    messages: List[ChatCompletionMessageParam]

1337
1338
1339
1340
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
    add_generation_prompt: bool = Field(
        default=True,
        description=
        ("If true, the generation prompt will be added to the chat template. "
         "This is a parameter used by chat template in tokenizer config of the "
         "model."),
    )
    continue_final_message: bool = Field(
        default=False,
        description=
        ("If this is set, the chat will be formatted so that the final "
         "message in the chat is open-ended, without any EOS tokens. The "
         "model will continue this message rather than starting a new one. "
         "This allows you to \"prefill\" part of the model's response for it. "
         "Cannot be used at the same time as `add_generation_prompt`."),
    )
    add_special_tokens: bool = Field(
        default=False,
        description=(
            "If true, special tokens (e.g. BOS) will be added to the prompt "
            "on top of what is added by the chat template. "
            "For most models, the chat template takes care of adding the "
            "special tokens so this should be set to false (as is the "
            "default)."),
    )
    chat_template: Optional[str] = Field(
        default=None,
        description=(
            "A Jinja template to use for this conversion. "
            "As of transformers v4.44, default chat template is no longer "
            "allowed, so you must provide a chat template if the tokenizer "
            "does not define one."),
    )
    chat_template_kwargs: Optional[Dict[str, Any]] = Field(
        default=None,
        description=("Additional kwargs to pass to the template renderer. "
                     "Will be accessible by the chat template."),
    )
1375

1376
1377
1378
1379
1380
1381
1382
1383
1384
    @model_validator(mode="before")
    @classmethod
    def check_generation_prompt(cls, data):
        if data.get("continue_final_message") and data.get(
                "add_generation_prompt"):
            raise ValueError("Cannot set both `continue_final_message` and "
                             "`add_generation_prompt` to True.")
        return data

1385
1386

TokenizeRequest = Union[TokenizeCompletionRequest, TokenizeChatRequest]
1387
1388
1389
1390
1391


class TokenizeResponse(OpenAIBaseModel):
    count: int
    max_model_len: int
1392
    tokens: List[int]
1393
1394
1395
1396
1397
1398
1399
1400
1401


class DetokenizeRequest(OpenAIBaseModel):
    model: str
    tokens: List[int]


class DetokenizeResponse(OpenAIBaseModel):
    prompt: str
1402
1403
1404
1405
1406
1407
1408
1409
1410
1411


class LoadLoraAdapterRequest(BaseModel):
    lora_name: str
    lora_path: str


class UnloadLoraAdapterRequest(BaseModel):
    lora_name: str
    lora_int_id: Optional[int] = Field(default=None)