"vllm/vscode:/vscode.git/clone" did not exist on "a4113b035cd4ee90ec02de0658b75d47f0007ede"
protocol.py 56.7 KB
Newer Older
1
2
# SPDX-License-Identifier: Apache-2.0

3
4
# Adapted from
# https://github.com/lm-sys/FastChat/blob/168ccc29d3f7edc50823016105c024fe2282732a/fastchat/protocol/openai_api_protocol.py
5
import re
Zhuohan Li's avatar
Zhuohan Li committed
6
import time
7
from argparse import Namespace
8
from typing import Any, ClassVar, Dict, List, Literal, Optional, Set, Union
Zhuohan Li's avatar
Zhuohan Li committed
9

10
import torch
11
from fastapi import UploadFile
12
13
from pydantic import (BaseModel, ConfigDict, Field, TypeAdapter,
                      ValidationInfo, field_validator, model_validator)
14
from typing_extensions import Annotated, TypeAlias
Zhuohan Li's avatar
Zhuohan Li committed
15

16
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
17
from vllm.logger import init_logger
18
from vllm.pooling_params import PoolingParams
19
20
from vllm.sampling_params import (BeamSearchParams, GuidedDecodingParams,
                                  RequestOutputKind, SamplingParams)
21
from vllm.sequence import Logprob
22
from vllm.utils import random_uuid, resolve_obj_by_qualname
23

24
25
logger = init_logger(__name__)

26
27
28
# torch is mocked during docs generation,
# so we have to provide the values as literals
_MOCK_LONG_INFO = Namespace(min=-9223372036854775808, max=9223372036854775807)
29
_LONG_INFO: Union["torch.iinfo", Namespace]
30
31
32
33
34
35
36
37
38
39
40
41
42
43

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
44

45
class OpenAIBaseModel(BaseModel):
46
47
48
    # OpenAI API does allow extra fields
    model_config = ConfigDict(extra="allow")

49
50
51
    # Cache class field names
    field_names: ClassVar[Optional[Set[str]]] = None

52
    @model_validator(mode="wrap")
53
    @classmethod
54
55
56
57
    def __log_extra_fields__(cls, data, handler):
        result = handler(data)
        if not isinstance(data, dict):
            return result
58
59
        field_names = cls.field_names
        if field_names is None:
60
61
62
63
            # 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)
64
65
66
67
68
69
70
71
72
73
                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)
74
        return result
75
76
77


class ErrorResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
78
79
80
81
    object: str = "error"
    message: str
    type: str
    param: Optional[str] = None
82
    code: int
Zhuohan Li's avatar
Zhuohan Li committed
83
84


85
class ModelPermission(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
86
87
88
89
90
91
92
93
94
95
96
    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
97
    is_blocking: bool = False
Zhuohan Li's avatar
Zhuohan Li committed
98
99


100
class ModelCard(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
101
102
103
    id: str
    object: str = "model"
    created: int = Field(default_factory=lambda: int(time.time()))
Woosuk Kwon's avatar
Woosuk Kwon committed
104
    owned_by: str = "vllm"
Zhuohan Li's avatar
Zhuohan Li committed
105
106
    root: Optional[str] = None
    parent: Optional[str] = None
107
    max_model_len: Optional[int] = None
Zhuohan Li's avatar
Zhuohan Li committed
108
109
110
    permission: List[ModelPermission] = Field(default_factory=list)


111
class ModelList(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
112
113
114
115
    object: str = "list"
    data: List[ModelCard] = Field(default_factory=list)


116
117
118
119
class PromptTokenUsageInfo(OpenAIBaseModel):
    cached_tokens: Optional[int] = None


120
class UsageInfo(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
121
122
123
    prompt_tokens: int = 0
    total_tokens: int = 0
    completion_tokens: Optional[int] = 0
124
    prompt_tokens_details: Optional[PromptTokenUsageInfo] = None
Zhuohan Li's avatar
Zhuohan Li committed
125
126


127
128
129
130
131
class RequestResponseMetadata(BaseModel):
    request_id: str
    final_usage_info: Optional[UsageInfo] = None


132
133
134
135
136
137
138
139
140
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


141
class ResponseFormat(OpenAIBaseModel):
142
143
144
    # type must be "json_schema", "json_object" or "text"
    type: Literal["text", "json_object", "json_schema"]
    json_schema: Optional[JsonSchemaResponseFormat] = None
145
146


147
class StreamOptions(OpenAIBaseModel):
148
    include_usage: Optional[bool] = True
149
    continuous_usage_stats: Optional[bool] = False
150
151


152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
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"


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
208
209
210
211
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


212
class ChatCompletionRequest(OpenAIBaseModel):
213
214
    # Ordered by official OpenAI API documentation
    # https://platform.openai.com/docs/api-reference/chat/create
215
    messages: List[ChatCompletionMessageParam]
216
217
218
219
    model: str
    frequency_penalty: Optional[float] = 0.0
    logit_bias: Optional[Dict[str, float]] = None
    logprobs: Optional[bool] = False
220
    top_logprobs: Optional[int] = 0
221
222
223
224
225
226
    # 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
227
228
229
    n: Optional[int] = 1
    presence_penalty: Optional[float] = 0.0
    response_format: Optional[ResponseFormat] = None
230
    seed: Optional[int] = Field(None, ge=_LONG_INFO.min, le=_LONG_INFO.max)
231
    stop: Optional[Union[str, List[str]]] = Field(default_factory=list)
Zhuohan Li's avatar
Zhuohan Li committed
232
    stream: Optional[bool] = False
233
    stream_options: Optional[StreamOptions] = None
234
235
    temperature: Optional[float] = None
    top_p: Optional[float] = None
236
    tools: Optional[List[ChatCompletionToolsParam]] = None
237
    tool_choice: Optional[Union[Literal["none"], Literal["auto"],
238
                                ChatCompletionNamedToolChoiceParam]] = "none"
239
240
241

    # 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
242
    user: Optional[str] = None
243
244

    # doc: begin-chat-completion-sampling-params
245
    best_of: Optional[int] = None
246
    use_beam_search: bool = False
247
248
249
    top_k: Optional[int] = None
    min_p: Optional[float] = None
    repetition_penalty: Optional[float] = None
250
    length_penalty: float = 1.0
251
    stop_token_ids: Optional[List[int]] = Field(default_factory=list)
252
253
254
255
256
257
    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
258
    prompt_logprobs: Optional[int] = None
259
260
261
    # doc: end-chat-completion-sampling-params

    # doc: begin-chat-completion-extra-params
262
    echo: bool = Field(
263
264
265
266
267
        default=False,
        description=(
            "If true, the new message will be prepended with the last message "
            "if they belong to the same role."),
    )
268
    add_generation_prompt: bool = Field(
269
270
271
272
273
274
        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."),
    )
275
276
277
278
279
280
281
282
283
    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`."),
    )
284
    add_special_tokens: bool = Field(
285
286
287
288
289
        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 "
290
            "special tokens so this should be set to false (as is the "
291
292
            "default)."),
    )
293
294
295
296
297
298
299
300
301
302
303
304
305
    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. "
306
307
308
            "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."),
309
310
311
312
313
314
    )
    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."),
    )
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
    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."),
    )
334
335
336
337
338
339
    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'"))
340
341
342
343
344
    guided_whitespace_pattern: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default whitespace pattern "
            "for guided json decoding."))
345
346
347
348
349
350
    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."))
351
352
353
354
355
356
    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."))
357
358
359
360
361
362
363
364
365
366
367
    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'}}."))
368
369

    # doc: end-chat-completion-extra-params
Zhuohan Li's avatar
Zhuohan Li committed
370

371
372
373
374
375
376
377
378
379
380
381
382
383
384
    # 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:
385
386
        # TODO(#9845): remove max_tokens when field is removed from OpenAI API
        max_tokens = self.max_completion_tokens or self.max_tokens
387

388
389
        if default_sampling_params is None:
            default_sampling_params = {}
390
        n = self.n if self.n is not None else 1
391

392
393
394
395
396
397
        # 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)

398
399
400
        if (temperature := self.temperature) is None:
            temperature = default_sampling_params.get(
                "temperature", self._DEFAULT_SAMPLING_PARAMS["temperature"])
401
402
403
404
405
406

        return BeamSearchParams(
            beam_width=n,
            max_tokens=max_tokens,
            ignore_eos=self.ignore_eos,
            temperature=temperature,
407
            length_penalty=self.length_penalty,
408
            include_stop_str_in_output=self.include_stop_str_in_output)
409

410
    def to_sampling_params(
411
412
413
414
            self,
            default_max_tokens: int,
            logits_processor_pattern: Optional[str],
            default_sampling_params: Optional[dict] = None) -> SamplingParams:
415
416
        # TODO(#9845): remove max_tokens when field is removed from OpenAI API
        max_tokens = self.max_completion_tokens or self.max_tokens
417

418
419
        if default_sampling_params is None:
            default_sampling_params = {}
420
421
422
423
424
425
426

        # 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)

427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
        # 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"])

446
447
448
449
        prompt_logprobs = self.prompt_logprobs
        if prompt_logprobs is None and self.echo:
            prompt_logprobs = self.top_logprobs

450
        guided_json_object = None
451
452
453
454
455
456
457
458
        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:
459
                    self.guided_decoding_backend = "xgrammar"
460
461
462
463
464
465
466
467
468

        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)
469

470
        return SamplingParams.from_optional(
471
            n=self.n,
472
            best_of=self.best_of,
473
474
            presence_penalty=self.presence_penalty,
            frequency_penalty=self.frequency_penalty,
475
476
477
478
479
            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
480
            seed=self.seed,
481
482
            stop=self.stop,
            stop_token_ids=self.stop_token_ids,
483
            logprobs=self.top_logprobs if self.logprobs else None,
484
            prompt_logprobs=prompt_logprobs,
485
            ignore_eos=self.ignore_eos,
486
            max_tokens=max_tokens,
487
            min_tokens=self.min_tokens,
488
489
            skip_special_tokens=self.skip_special_tokens,
            spaces_between_special_tokens=self.spaces_between_special_tokens,
490
491
            logits_processors=get_logits_processors(self.logits_processors,
                                                    logits_processor_pattern),
492
            include_stop_str_in_output=self.include_stop_str_in_output,
493
            truncate_prompt_tokens=self.truncate_prompt_tokens,
494
495
            output_kind=RequestOutputKind.DELTA if self.stream \
                else RequestOutputKind.FINAL_ONLY,
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
            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
516

517
    @model_validator(mode="before")
518
    @classmethod
519
520
    def validate_stream_options(cls, data):
        if data.get("stream_options") and not data.get("stream"):
521
            raise ValueError(
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
                "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
547

548
549
550
    @model_validator(mode="before")
    @classmethod
    def check_guided_decoding_count(cls, data):
551
552
553
        if isinstance(data, ValueError):
            raise data

554
555
556
557
558
        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
        ])
559
        # you can only use one kind of guided decoding
560
561
562
563
        if guide_count > 1:
            raise ValueError(
                "You can only use one kind of guided decoding "
                "('guided_json', 'guided_regex' or 'guided_choice').")
564
        # you can only either use guided decoding or tools, not both
565
566
        if guide_count > 1 and data.get("tool_choice",
                                        "none") not in ("none", "auto"):
567
568
569
570
571
572
            raise ValueError(
                "You can only either use guided decoding or tools, not both.")
        return data

    @model_validator(mode="before")
    @classmethod
573
574
575
576
    def check_tool_usage(cls, data):

        # if "tool_choice" is not specified but tools are provided,
        # default to "auto" tool_choice
577
        if "tool_choice" not in data and data.get("tools"):
578
579
            data["tool_choice"] = "auto"

580
581
582
583
584
585
        # 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

586
587
588
589
        # if "tool_choice" is specified -- validation
        if "tool_choice" in data:

            # ensure that if "tool choice" is specified, tools are present
590
591
592
            if "tools" not in data or data["tools"] is None:
                raise ValueError(
                    "When using `tool_choice`, `tools` must be set.")
593
594
595
596
597
598

            # 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(
599
600
                    "`tool_choice` must either be a named tool, \"auto\", "
                    "or \"none\".")
601
602
603
604
605

            # ensure that if "tool_choice" is specified as an object,
            # it matches a valid tool
            if isinstance(data["tool_choice"], dict):
                valid_tool = False
606
                specified_function = data["tool_choice"].get("function")
607
608
                if not specified_function:
                    raise ValueError(
609
610
                        "Expected field `function` in `tool_choice`."
                        " Correct usage: `{\"type\": \"function\","
611
                        " \"function\": {\"name\": \"my_function\"}}`")
612
                specified_function_name = specified_function.get("name")
613
614
                if not specified_function_name:
                    raise ValueError(
615
616
                        "Expected field `name` in `function` in `tool_choice`."
                        "Correct usage: `{\"type\": \"function\", "
617
618
619
620
621
622
623
624
625
                        "\"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`")
626
627
        return data

628
629
630
631
632
633
634
635
636
    @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
637

638
class CompletionRequest(OpenAIBaseModel):
639
640
    # Ordered by official OpenAI API documentation
    # https://platform.openai.com/docs/api-reference/completions/create
Zhuohan Li's avatar
Zhuohan Li committed
641
    model: str
642
    prompt: Union[List[int], List[List[int]], str, List[str]]
643
    best_of: Optional[int] = None
Zhuohan Li's avatar
Zhuohan Li committed
644
645
646
    echo: Optional[bool] = False
    frequency_penalty: Optional[float] = 0.0
    logit_bias: Optional[Dict[str, float]] = None
647
648
    logprobs: Optional[int] = None
    max_tokens: Optional[int] = 16
649
    n: int = 1
650
    presence_penalty: Optional[float] = 0.0
651
    seed: Optional[int] = Field(None, ge=_LONG_INFO.min, le=_LONG_INFO.max)
652
653
    stop: Optional[Union[str, List[str]]] = Field(default_factory=list)
    stream: Optional[bool] = False
654
    stream_options: Optional[StreamOptions] = None
655
    suffix: Optional[str] = None
656
657
    temperature: Optional[float] = None
    top_p: Optional[float] = None
Zhuohan Li's avatar
Zhuohan Li committed
658
    user: Optional[str] = None
659
660

    # doc: begin-completion-sampling-params
661
    use_beam_search: bool = False
662
663
664
    top_k: Optional[int] = None
    min_p: Optional[float] = None
    repetition_penalty: Optional[float] = None
665
    length_penalty: float = 1.0
666
    stop_token_ids: Optional[List[int]] = Field(default_factory=list)
667
668
669
670
671
    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
672
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None
673
    allowed_token_ids: Optional[List[int]] = None
674
    prompt_logprobs: Optional[int] = None
675
676
677
    # doc: end-completion-sampling-params

    # doc: begin-completion-extra-params
678
679
    add_special_tokens: bool = Field(
        default=True,
680
        description=(
681
682
            "If true (the default), special tokens (e.g. BOS) will be added to "
            "the prompt."),
683
684
685
686
687
    )
    response_format: Optional[ResponseFormat] = Field(
        default=None,
        description=
        ("Similar to chat completion, this parameter specifies the format of "
688
689
         "output. Only {'type': 'json_object'}, {'type': 'json_schema'} or "
         "{'type': 'text' } is supported."),
690
691
692
    )
    guided_json: Optional[Union[str, dict, BaseModel]] = Field(
        default=None,
693
        description="If specified, the output will follow the JSON schema.",
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
    )
    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."),
    )
710
711
712
713
714
715
    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'"))
716
717
718
719
720
    guided_whitespace_pattern: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default whitespace pattern "
            "for guided json decoding."))
721
722
723
724
725
726
    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."))
727
728
729
730
731
732
733
734
735
736
737
    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'}}."))
738
739

    # doc: end-completion-extra-params
Zhuohan Li's avatar
Zhuohan Li committed
740

741
742
743
744
745
746
747
748
749
750
751
752
753
754
    # 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:
755
756
        max_tokens = self.max_tokens

757
758
        if default_sampling_params is None:
            default_sampling_params = {}
759
        n = self.n if self.n is not None else 1
760

761
762
763
764
765
766
        # 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)

767
768
        if (temperature := self.temperature) is None:
            temperature = default_sampling_params.get("temperature", 1.0)
769
770
771
772
773
774

        return BeamSearchParams(
            beam_width=n,
            max_tokens=max_tokens,
            ignore_eos=self.ignore_eos,
            temperature=temperature,
775
            length_penalty=self.length_penalty,
776
            include_stop_str_in_output=self.include_stop_str_in_output)
777

778
    def to_sampling_params(
779
780
781
782
            self,
            default_max_tokens: int,
            logits_processor_pattern: Optional[str],
            default_sampling_params: Optional[dict] = None) -> SamplingParams:
783
784
        max_tokens = self.max_tokens

785
786
        if default_sampling_params is None:
            default_sampling_params = {}
787
788
789
790
791
792
793

        # 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)

794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
        # 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"])

813
814
815
816
        prompt_logprobs = self.prompt_logprobs
        if prompt_logprobs is None and self.echo:
            prompt_logprobs = self.logprobs

817
818
        echo_without_generation = self.echo and self.max_tokens == 0

819
820
821
822
823
824
825
826
827
828
829
830
831
        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)
832

833
        return SamplingParams.from_optional(
834
835
836
837
            n=self.n,
            best_of=self.best_of,
            presence_penalty=self.presence_penalty,
            frequency_penalty=self.frequency_penalty,
838
839
840
841
842
            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
843
            seed=self.seed,
844
845
            stop=self.stop,
            stop_token_ids=self.stop_token_ids,
846
            logprobs=self.logprobs,
847
            ignore_eos=self.ignore_eos,
848
            max_tokens=max_tokens if not echo_without_generation else 1,
849
            min_tokens=self.min_tokens,
850
            prompt_logprobs=prompt_logprobs,
851
            skip_special_tokens=self.skip_special_tokens,
852
            spaces_between_special_tokens=self.spaces_between_special_tokens,
853
            include_stop_str_in_output=self.include_stop_str_in_output,
854
855
            logits_processors=get_logits_processors(self.logits_processors,
                                                    logits_processor_pattern),
856
            truncate_prompt_tokens=self.truncate_prompt_tokens,
857
858
            output_kind=RequestOutputKind.DELTA if self.stream \
                else RequestOutputKind.FINAL_ONLY,
859
860
861
            guided_decoding=guided_decoding,
            logit_bias=self.logit_bias,
            allowed_token_ids=self.allowed_token_ids)
862

863
864
865
866
867
868
869
870
871
872
873
874
875
876
    @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

877
878
879
    @model_validator(mode="before")
    @classmethod
    def check_logprobs(cls, data):
880
881
882
883
884
885
886
887
888
889
890
        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.")

891
892
        return data

893
894
895
896
897
    @model_validator(mode="before")
    @classmethod
    def validate_stream_options(cls, data):
        if data.get("stream_options") and not data.get("stream"):
            raise ValueError(
898
899
                "Stream options can only be defined when `stream=True`.")

900
901
        return data

Zhuohan Li's avatar
Zhuohan Li committed
902

903
class EmbeddingCompletionRequest(OpenAIBaseModel):
904
905
906
907
    # 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]]
908
    encoding_format: Literal["float", "base64"] = "float"
909
910
    dimensions: Optional[int] = None
    user: Optional[str] = None
911
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None
912
913
914
915
916

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

917
    # doc: begin-embedding-extra-params
918
919
920
921
922
923
    add_special_tokens: bool = Field(
        default=True,
        description=(
            "If true (the default), special tokens (e.g. BOS) will be added to "
            "the prompt."),
    )
924
925
926
927
928
929
930
931
932
    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

933
934
935
936
    def to_pooling_params(self):
        return PoolingParams(additional_data=self.additional_data)


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
992
993
994
995
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]

996
997
998
999
PoolingCompletionRequest = EmbeddingCompletionRequest
PoolingChatRequest = EmbeddingChatRequest
PoolingRequest = Union[PoolingCompletionRequest, PoolingChatRequest]

1000

1001
1002
1003
1004
1005
1006
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

1007
    # doc: begin-score-pooling-params
1008
    additional_data: Optional[Any] = None
1009
    # doc: end-score-pooling-params
1010

1011
    # doc: begin-score-extra-params
1012
1013
1014
1015
1016
1017
1018
    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."))

1019
1020
    # doc: end-score-extra-params

1021
1022
1023
1024
    def to_pooling_params(self):
        return PoolingParams(additional_data=self.additional_data)


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
1067
1068
1069
1070
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]


1071
class CompletionLogProbs(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1072
1073
1074
    text_offset: List[int] = Field(default_factory=list)
    token_logprobs: List[Optional[float]] = Field(default_factory=list)
    tokens: List[str] = Field(default_factory=list)
1075
1076
    top_logprobs: List[Optional[Dict[str,
                                     float]]] = Field(default_factory=list)
Zhuohan Li's avatar
Zhuohan Li committed
1077
1078


1079
class CompletionResponseChoice(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1080
1081
    index: int
    text: str
1082
    logprobs: Optional[CompletionLogProbs] = None
1083
1084
    finish_reason: Optional[str] = None
    stop_reason: Optional[Union[int, str]] = Field(
1085
1086
1087
1088
1089
1090
        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"),
    )
1091
    prompt_logprobs: Optional[List[Optional[Dict[int, Logprob]]]] = None
Zhuohan Li's avatar
Zhuohan Li committed
1092
1093


1094
class CompletionResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1095
1096
1097
1098
1099
1100
1101
1102
    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


1103
class CompletionResponseStreamChoice(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1104
1105
    index: int
    text: str
1106
    logprobs: Optional[CompletionLogProbs] = None
1107
1108
    finish_reason: Optional[str] = None
    stop_reason: Optional[Union[int, str]] = Field(
1109
1110
1111
1112
1113
1114
        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
1115
1116


1117
class CompletionStreamResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
1118
1119
1120
1121
1122
    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]
1123
    usage: Optional[UsageInfo] = Field(default=None)
1124
1125


1126
class EmbeddingResponseData(OpenAIBaseModel):
1127
1128
    index: int
    object: str = "embedding"
1129
    embedding: Union[List[float], str]
1130
1131


1132
class EmbeddingResponse(OpenAIBaseModel):
1133
    id: str = Field(default_factory=lambda: f"embd-{random_uuid()}")
1134
1135
1136
1137
1138
1139
1140
    object: str = "list"
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    data: List[EmbeddingResponseData]
    usage: UsageInfo


1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
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


1156
1157
1158
class ScoreResponseData(OpenAIBaseModel):
    index: int
    object: str = "score"
1159
    score: float
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170


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


1171
1172
1173
1174
1175
1176
1177
1178
1179
1180
1181
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


1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
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


1207
class ChatMessage(OpenAIBaseModel):
1208
    role: str
1209
    reasoning_content: Optional[str] = None
1210
    content: Optional[str] = None
1211
    tool_calls: List[ToolCall] = Field(default_factory=list)
1212
1213


1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
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


1228
class ChatCompletionResponseChoice(OpenAIBaseModel):
1229
1230
    index: int
    message: ChatMessage
1231
    logprobs: Optional[ChatCompletionLogProbs] = None
1232
1233
1234
    # 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
1235
    stop_reason: Optional[Union[int, str]] = None
1236
1237


1238
class ChatCompletionResponse(OpenAIBaseModel):
1239
    id: str = Field(default_factory=lambda: f"chatcmpl-{random_uuid()}")
1240
    object: Literal["chat.completion"] = "chat.completion"
1241
1242
1243
1244
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[ChatCompletionResponseChoice]
    usage: UsageInfo
1245
    prompt_logprobs: Optional[List[Optional[Dict[int, Logprob]]]] = None
1246
1247


1248
class DeltaMessage(OpenAIBaseModel):
1249
1250
    role: Optional[str] = None
    content: Optional[str] = None
1251
    reasoning_content: Optional[str] = None
1252
    tool_calls: List[DeltaToolCall] = Field(default_factory=list)
1253
1254


1255
class ChatCompletionResponseStreamChoice(OpenAIBaseModel):
1256
1257
    index: int
    delta: DeltaMessage
1258
    logprobs: Optional[ChatCompletionLogProbs] = None
1259
    finish_reason: Optional[str] = None
1260
    stop_reason: Optional[Union[int, str]] = None
1261
1262


1263
class ChatCompletionStreamResponse(OpenAIBaseModel):
1264
    id: str = Field(default_factory=lambda: f"chatcmpl-{random_uuid()}")
1265
    object: Literal["chat.completion.chunk"] = "chat.completion.chunk"
1266
1267
1268
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[ChatCompletionResponseStreamChoice]
1269
    usage: Optional[UsageInfo] = Field(default=None)
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290


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

1291
    # The parameters of the request.
1292
    body: Union[ChatCompletionRequest, EmbeddingRequest, ScoreRequest]
1293

1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
    @field_validator('body', mode='plain')
    @classmethod
    def check_type_for_url(cls, value: Any, info: ValidationInfo):
        # Use url to disambiguate models
        url = info.data['url']
        if url == "/v1/chat/completions":
            return ChatCompletionRequest.model_validate(value)
        if url == "/v1/embeddings":
            return TypeAdapter(EmbeddingRequest).validate_python(value)
        if url == "/v1/score":
            return ScoreRequest.model_validate(value)
        return TypeAdapter(Union[ChatCompletionRequest, EmbeddingRequest,
                                 ScoreRequest]).validate_python(value)

1308

1309
1310
1311
1312
1313
1314
1315
1316
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.
1317
1318
    body: Optional[Union[ChatCompletionResponse, EmbeddingResponse,
                         ScoreResponse]] = None
1319
1320


1321
1322
1323
1324
1325
1326
1327
1328
1329
1330
1331
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

1332
    response: Optional[BatchResponseData]
1333
1334
1335
1336

    # For requests that failed with a non-HTTP error, this will contain more
    # information on the cause of the failure.
    error: Optional[Any]
1337
1338


1339
1340
1341
1342
class TokenizeCompletionRequest(OpenAIBaseModel):
    model: str
    prompt: str

1343
1344
1345
1346
1347
1348
    add_special_tokens: bool = Field(
        default=True,
        description=(
            "If true (the default), special tokens (e.g. BOS) will be added to "
            "the prompt."),
    )
1349
1350
1351
1352
1353
1354


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

1355
1356
1357
1358
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374
1375
1376
1377
1378
1379
1380
1381
1382
1383
1384
1385
1386
1387
1388
1389
1390
1391
1392
    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."),
    )
1393

1394
1395
1396
1397
1398
1399
1400
1401
1402
    @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

1403
1404

TokenizeRequest = Union[TokenizeCompletionRequest, TokenizeChatRequest]
1405
1406
1407
1408
1409


class TokenizeResponse(OpenAIBaseModel):
    count: int
    max_model_len: int
1410
    tokens: List[int]
1411
1412
1413
1414
1415
1416
1417
1418
1419


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


class DetokenizeResponse(OpenAIBaseModel):
    prompt: str
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429


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


class UnloadLoraAdapterRequest(BaseModel):
    lora_name: str
    lora_int_id: Optional[int] = Field(default=None)
1430
1431
1432
1433
1434
1435
1436
1437
1438
1439
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452
1453
1454
1455
1456
1457
1458
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485
1486
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550
1551
1552
1553
1554
1555
1556
1557
1558
1559
1560
1561
1562
1563
1564
1565
1566
1567
1568
1569
1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589


## Protocols for Audio
AudioResponseFormat: TypeAlias = Literal["json", "text", "srt", "verbose_json",
                                         "vtt"]


class TranscriptionRequest(OpenAIBaseModel):
    # Ordered by official OpenAI API documentation
    #https://platform.openai.com/docs/api-reference/audio/createTranscription

    file: UploadFile
    """
    The audio file object (not file name) to transcribe, in one of these
    formats: flac, mp3, mp4, mpeg, mpga, m4a, ogg, wav, or webm.
    """

    model: str
    """ID of the model to use.
    """

    language: Optional[str] = None
    """The language of the input audio.

    Supplying the input language in
    [ISO-639-1](https://en.wikipedia.org/wiki/List_of_ISO_639-1_codes) format
    will improve accuracy and latency.
    """

    prompt: str = Field(default="")
    """An optional text to guide the model's style or continue a previous audio
    segment.

    The [prompt](https://platform.openai.com/docs/guides/speech-to-text#prompting)
    should match the audio language.
    """

    response_format: AudioResponseFormat = Field(default="json")
    """
    The format of the output, in one of these options: `json`, `text`, `srt`,
    `verbose_json`, or `vtt`.
    """

    ## TODO (varun) : Support if set to 0, certain thresholds are met !!
    temperature: float = Field(default=0.0)
    """The sampling temperature, between 0 and 1.

    Higher values like 0.8 will make the output more random, while lower values
    like 0.2 will make it more focused / deterministic. If set to 0, the model
    will use [log probability](https://en.wikipedia.org/wiki/Log_probability)
    to automatically increase the temperature until certain thresholds are hit.
    """

    timestamp_granularities: List[Literal["word", "segment"]] = Field(
        alias="timestamp_granularities[]", default=[])
    """The timestamp granularities to populate for this transcription.

    `response_format` must be set `verbose_json` to use timestamp granularities.
    Either or both of these options are supported: `word`, or `segment`. Note:
    There is no additional latency for segment timestamps, but generating word
    timestamps incurs additional latency.
    """

    # Default sampling parameters for transcription requests.
    _DEFAULT_SAMPLING_PARAMS: dict = {
        "temperature": 0,
    }

    def to_sampling_params(
            self,
            default_max_tokens: int,
            default_sampling_params: Optional[dict] = None) -> SamplingParams:
        # TODO(#9845): remove max_tokens when field is removed from OpenAI API
        max_tokens = default_max_tokens

        if default_sampling_params is None:
            default_sampling_params = {}
        # Default parameters
        if (temperature := self.temperature) is None:
            temperature = default_sampling_params.get(
                "temperature", self._DEFAULT_SAMPLING_PARAMS["temperature"])

        return SamplingParams.from_optional(temperature=temperature,
                                            max_tokens=max_tokens)


# Transcription response objects
class TranscriptionResponse(OpenAIBaseModel):
    text: str
    """The transcribed text."""


class TranscriptionWord(OpenAIBaseModel):
    end: float
    """End time of the word in seconds."""

    start: float
    """Start time of the word in seconds."""

    word: str
    """The text content of the word."""


class TranscriptionSegment(OpenAIBaseModel):
    id: int
    """Unique identifier of the segment."""

    avg_logprob: float
    """Average logprob of the segment.

    If the value is lower than -1, consider the logprobs failed.
    """

    compression_ratio: float
    """Compression ratio of the segment.

    If the value is greater than 2.4, consider the compression failed.
    """

    end: float
    """End time of the segment in seconds."""

    no_speech_prob: float
    """Probability of no speech in the segment.

    If the value is higher than 1.0 and the `avg_logprob` is below -1, consider
    this segment silent.
    """

    seek: int
    """Seek offset of the segment."""

    start: float
    """Start time of the segment in seconds."""

    temperature: float
    """Temperature parameter used for generating the segment."""

    text: str
    """Text content of the segment."""

    tokens: List[int]
    """Array of token IDs for the text content."""


class TranscriptionResponseVerbose(OpenAIBaseModel):
    duration: str
    """The duration of the input audio."""

    language: str
    """The language of the input audio."""

    text: str
    """The transcribed text."""

    segments: Optional[List[TranscriptionSegment]] = None
    """Segments of the transcribed text and their corresponding details."""

    words: Optional[List[TranscriptionWord]] = None
    """Extracted words and their corresponding timestamps."""