protocol.py 26.8 KB
Newer Older
1
2
# Adapted from
# https://github.com/lm-sys/FastChat/blob/168ccc29d3f7edc50823016105c024fe2282732a/fastchat/protocol/openai_api_protocol.py
Zhuohan Li's avatar
Zhuohan Li committed
3
import time
4
from argparse import Namespace
5
from typing import Any, Dict, List, Literal, Optional, Union
Zhuohan Li's avatar
Zhuohan Li committed
6

7
import torch
8
from pydantic import BaseModel, ConfigDict, Field, model_validator
9
from typing_extensions import Annotated
Zhuohan Li's avatar
Zhuohan Li committed
10

11
from vllm.entrypoints.chat_utils import ChatCompletionMessageParam
12
from vllm.entrypoints.openai.logits_processors import get_logits_processors
13
from vllm.pooling_params import PoolingParams
14
from vllm.sampling_params import LogitsProcessor, SamplingParams
15
from vllm.sequence import Logprob
16
from vllm.transformers_utils.tokenizer import AnyTokenizer
17
from vllm.utils import random_uuid
18

19
20
21
# torch is mocked during docs generation,
# so we have to provide the values as literals
_MOCK_LONG_INFO = Namespace(min=-9223372036854775808, max=9223372036854775807)
22
_LONG_INFO: Union["torch.iinfo", Namespace]
23
24
25
26
27
28
29
30
31
32
33
34
35
36

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
37

38
39
40
41
42
43
class OpenAIBaseModel(BaseModel):
    # OpenAI API does not allow extra fields
    model_config = ConfigDict(extra="forbid")


class ErrorResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
44
45
46
47
    object: str = "error"
    message: str
    type: str
    param: Optional[str] = None
48
    code: int
Zhuohan Li's avatar
Zhuohan Li committed
49
50


51
class ModelPermission(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
52
53
54
55
56
57
58
59
60
61
62
    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
63
    is_blocking: bool = False
Zhuohan Li's avatar
Zhuohan Li committed
64
65


66
class ModelCard(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
67
68
69
    id: str
    object: str = "model"
    created: int = Field(default_factory=lambda: int(time.time()))
Woosuk Kwon's avatar
Woosuk Kwon committed
70
    owned_by: str = "vllm"
Zhuohan Li's avatar
Zhuohan Li committed
71
72
    root: Optional[str] = None
    parent: Optional[str] = None
73
    max_model_len: Optional[int] = None
Zhuohan Li's avatar
Zhuohan Li committed
74
75
76
    permission: List[ModelPermission] = Field(default_factory=list)


77
class ModelList(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
78
79
80
81
    object: str = "list"
    data: List[ModelCard] = Field(default_factory=list)


82
class UsageInfo(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
83
84
85
86
87
    prompt_tokens: int = 0
    total_tokens: int = 0
    completion_tokens: Optional[int] = 0


88
89
90
91
92
93
94
95
96
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


97
class ResponseFormat(OpenAIBaseModel):
98
99
100
    # type must be "json_schema", "json_object" or "text"
    type: Literal["text", "json_object", "json_schema"]
    json_schema: Optional[JsonSchemaResponseFormat] = None
101
102


103
class StreamOptions(OpenAIBaseModel):
104
105
    include_usage: Optional[bool] = True
    continuous_usage_stats: Optional[bool] = True
106
107


108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
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"


128
class ChatCompletionRequest(OpenAIBaseModel):
129
130
    # Ordered by official OpenAI API documentation
    # https://platform.openai.com/docs/api-reference/chat/create
131
    messages: List[ChatCompletionMessageParam]
132
133
134
135
    model: str
    frequency_penalty: Optional[float] = 0.0
    logit_bias: Optional[Dict[str, float]] = None
    logprobs: Optional[bool] = False
136
    top_logprobs: Optional[int] = 0
137
    max_tokens: Optional[int] = None
138
139
140
    n: Optional[int] = 1
    presence_penalty: Optional[float] = 0.0
    response_format: Optional[ResponseFormat] = None
141
    seed: Optional[int] = Field(None, ge=_LONG_INFO.min, le=_LONG_INFO.max)
142
    stop: Optional[Union[str, List[str]]] = Field(default_factory=list)
Zhuohan Li's avatar
Zhuohan Li committed
143
    stream: Optional[bool] = False
144
    stream_options: Optional[StreamOptions] = None
145
146
    temperature: Optional[float] = 0.7
    top_p: Optional[float] = 1.0
147
148
149
    tools: Optional[List[ChatCompletionToolsParam]] = None
    tool_choice: Optional[Union[Literal["none"],
                                ChatCompletionNamedToolChoiceParam]] = "none"
Zhuohan Li's avatar
Zhuohan Li committed
150
    user: Optional[str] = None
151
152

    # doc: begin-chat-completion-sampling-params
153
    best_of: Optional[int] = None
154
155
156
157
158
159
    use_beam_search: bool = False
    top_k: int = -1
    min_p: float = 0.0
    repetition_penalty: float = 1.0
    length_penalty: float = 1.0
    early_stopping: bool = False
160
    stop_token_ids: Optional[List[int]] = Field(default_factory=list)
161
162
163
164
165
166
    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
167
    prompt_logprobs: Optional[int] = None
168
169
170
    # doc: end-chat-completion-sampling-params

    # doc: begin-chat-completion-extra-params
171
    echo: bool = Field(
172
173
174
175
176
        default=False,
        description=(
            "If true, the new message will be prepended with the last message "
            "if they belong to the same role."),
    )
177
    add_generation_prompt: bool = Field(
178
179
180
181
182
183
        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."),
    )
184
    add_special_tokens: bool = Field(
185
186
187
188
189
        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 "
190
            "special tokens so this should be set to false (as is the "
191
192
            "default)."),
    )
193
194
195
196
197
198
199
200
201
202
203
204
205
    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. "
206
207
208
            "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."),
209
210
211
212
213
214
    )
    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."),
    )
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
    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."),
    )
234
235
236
237
238
239
    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'"))
240
241
242
243
244
    guided_whitespace_pattern: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default whitespace pattern "
            "for guided json decoding."))
245
246

    # doc: end-chat-completion-extra-params
Zhuohan Li's avatar
Zhuohan Li committed
247

248
    def to_sampling_params(
249
            self, tokenizer: AnyTokenizer,
250
251
252
253
254
            guided_decode_logits_processor: Optional[LogitsProcessor],
            default_max_tokens: int) -> SamplingParams:
        max_tokens = self.max_tokens
        if max_tokens is None:
            max_tokens = default_max_tokens
255

256
257
258
259
        prompt_logprobs = self.prompt_logprobs
        if prompt_logprobs is None and self.echo:
            prompt_logprobs = self.top_logprobs

260
        # We now allow logprobs being true without top_logrobs.
261
262
263
264
265
        logits_processors = get_logits_processors(
            logit_bias=self.logit_bias,
            allowed_token_ids=None,
            tokenizer=tokenizer,
        )
266
267
        if guided_decode_logits_processor:
            logits_processors.append(guided_decode_logits_processor)
268

269
        return SamplingParams.from_optional(
270
            n=self.n,
271
            best_of=self.best_of,
272
273
274
275
276
            presence_penalty=self.presence_penalty,
            frequency_penalty=self.frequency_penalty,
            repetition_penalty=self.repetition_penalty,
            temperature=self.temperature,
            top_p=self.top_p,
277
            top_k=self.top_k,
278
            min_p=self.min_p,
Nick Hill's avatar
Nick Hill committed
279
            seed=self.seed,
280
281
            stop=self.stop,
            stop_token_ids=self.stop_token_ids,
282
            logprobs=self.top_logprobs if self.logprobs else None,
283
            prompt_logprobs=prompt_logprobs,
284
            ignore_eos=self.ignore_eos,
285
            max_tokens=max_tokens,
286
            min_tokens=self.min_tokens,
287
            use_beam_search=self.use_beam_search,
288
            early_stopping=self.early_stopping,
289
290
            skip_special_tokens=self.skip_special_tokens,
            spaces_between_special_tokens=self.spaces_between_special_tokens,
291
292
            include_stop_str_in_output=self.include_stop_str_in_output,
            length_penalty=self.length_penalty,
293
            logits_processors=logits_processors,
294
            truncate_prompt_tokens=self.truncate_prompt_tokens,
295
296
        )

297
    @model_validator(mode="before")
298
    @classmethod
299
300
    def validate_stream_options(cls, data):
        if data.get("stream_options") and not data.get("stream"):
301
            raise ValueError(
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
                "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
327

328
329
330
331
332
333
334
335
    @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
        ])
336
        # you can only use one kind of guided decoding
337
338
339
340
        if guide_count > 1:
            raise ValueError(
                "You can only use one kind of guided decoding "
                "('guided_json', 'guided_regex' or 'guided_choice').")
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
        # you can only either use guided decoding or tools, not both
        if guide_count > 1 and "tool_choice" in data and data[
                "tool_choice"] != "none":
            raise ValueError(
                "You can only either use guided decoding or tools, not both.")
        return data

    @model_validator(mode="before")
    @classmethod
    def check_tool_choice(cls, data):
        if "tool_choice" in data and data["tool_choice"] != "none":
            if not isinstance(data["tool_choice"], dict):
                raise ValueError("Currently only named tools are supported.")
            if "tools" not in data or data["tools"] is None:
                raise ValueError(
                    "When using `tool_choice`, `tools` must be set.")
357
358
        return data

Zhuohan Li's avatar
Zhuohan Li committed
359

360
class CompletionRequest(OpenAIBaseModel):
361
362
    # Ordered by official OpenAI API documentation
    # https://platform.openai.com/docs/api-reference/completions/create
Zhuohan Li's avatar
Zhuohan Li committed
363
    model: str
364
    prompt: Union[List[int], List[List[int]], str, List[str]]
365
    best_of: Optional[int] = None
Zhuohan Li's avatar
Zhuohan Li committed
366
367
368
    echo: Optional[bool] = False
    frequency_penalty: Optional[float] = 0.0
    logit_bias: Optional[Dict[str, float]] = None
369
370
    logprobs: Optional[int] = None
    max_tokens: Optional[int] = 16
371
    n: int = 1
372
    presence_penalty: Optional[float] = 0.0
373
    seed: Optional[int] = Field(None, ge=_LONG_INFO.min, le=_LONG_INFO.max)
374
375
    stop: Optional[Union[str, List[str]]] = Field(default_factory=list)
    stream: Optional[bool] = False
376
    stream_options: Optional[StreamOptions] = None
377
378
379
    suffix: Optional[str] = None
    temperature: Optional[float] = 1.0
    top_p: Optional[float] = 1.0
Zhuohan Li's avatar
Zhuohan Li committed
380
    user: Optional[str] = None
381
382

    # doc: begin-completion-sampling-params
383
384
385
386
387
388
    use_beam_search: bool = False
    top_k: int = -1
    min_p: float = 0.0
    repetition_penalty: float = 1.0
    length_penalty: float = 1.0
    early_stopping: bool = False
389
    stop_token_ids: Optional[List[int]] = Field(default_factory=list)
390
391
392
393
394
    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
395
    truncate_prompt_tokens: Optional[Annotated[int, Field(ge=1)]] = None
396
    allowed_token_ids: Optional[List[int]] = None
397
    prompt_logprobs: Optional[int] = None
398
399
400
    # doc: end-completion-sampling-params

    # doc: begin-completion-extra-params
401
402
    add_special_tokens: bool = Field(
        default=True,
403
        description=(
404
405
            "If true (the default), special tokens (e.g. BOS) will be added to "
            "the prompt."),
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
    )
    response_format: Optional[ResponseFormat] = Field(
        default=None,
        description=
        ("Similar to chat completion, this parameter specifies the format of "
         "output. Only {'type': 'json_object'} or {'type': 'text' } is "
         "supported."),
    )
    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."),
    )
433
434
435
436
437
438
    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'"))
439
440
441
442
443
    guided_whitespace_pattern: Optional[str] = Field(
        default=None,
        description=(
            "If specified, will override the default whitespace pattern "
            "for guided json decoding."))
444
445

    # doc: end-completion-extra-params
Zhuohan Li's avatar
Zhuohan Li committed
446

447
    def to_sampling_params(
448
            self, tokenizer: AnyTokenizer,
449
450
451
452
453
454
            guided_decode_logits_processor: Optional[LogitsProcessor],
            default_max_tokens: int) -> SamplingParams:
        max_tokens = self.max_tokens
        if max_tokens is None:
            max_tokens = default_max_tokens

455
456
457
458
        prompt_logprobs = self.prompt_logprobs
        if prompt_logprobs is None and self.echo:
            prompt_logprobs = self.logprobs

459
460
        echo_without_generation = self.echo and self.max_tokens == 0

461
462
463
464
465
        logits_processors = get_logits_processors(
            logit_bias=self.logit_bias,
            allowed_token_ids=self.allowed_token_ids,
            tokenizer=tokenizer,
        )
466
467
        if guided_decode_logits_processor:
            logits_processors.append(guided_decode_logits_processor)
468

469
        return SamplingParams.from_optional(
470
471
472
473
474
475
476
477
478
            n=self.n,
            best_of=self.best_of,
            presence_penalty=self.presence_penalty,
            frequency_penalty=self.frequency_penalty,
            repetition_penalty=self.repetition_penalty,
            temperature=self.temperature,
            top_p=self.top_p,
            top_k=self.top_k,
            min_p=self.min_p,
Nick Hill's avatar
Nick Hill committed
479
            seed=self.seed,
480
481
            stop=self.stop,
            stop_token_ids=self.stop_token_ids,
482
            logprobs=self.logprobs,
483
            ignore_eos=self.ignore_eos,
484
            max_tokens=max_tokens if not echo_without_generation else 1,
485
            min_tokens=self.min_tokens,
486
            use_beam_search=self.use_beam_search,
487
            early_stopping=self.early_stopping,
488
            prompt_logprobs=prompt_logprobs,
489
            skip_special_tokens=self.skip_special_tokens,
490
            spaces_between_special_tokens=self.spaces_between_special_tokens,
491
492
            include_stop_str_in_output=self.include_stop_str_in_output,
            length_penalty=self.length_penalty,
493
            logits_processors=logits_processors,
494
            truncate_prompt_tokens=self.truncate_prompt_tokens,
495
496
        )

497
498
499
500
501
502
503
504
505
506
507
508
509
510
    @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

511
512
513
    @model_validator(mode="before")
    @classmethod
    def check_logprobs(cls, data):
514
515
516
517
518
519
520
521
522
523
524
        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.")

525
526
        return data

527
528
529
530
531
    @model_validator(mode="before")
    @classmethod
    def validate_stream_options(cls, data):
        if data.get("stream_options") and not data.get("stream"):
            raise ValueError(
532
533
                "Stream options can only be defined when `stream=True`.")

534
535
        return data

Zhuohan Li's avatar
Zhuohan Li committed
536

537
class EmbeddingRequest(OpenAIBaseModel):
538
539
540
541
    # 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]]
542
    encoding_format: Literal["float", "base64"] = "float"
543
544
545
546
547
548
549
550
551
552
553
554
    dimensions: Optional[int] = None
    user: Optional[str] = None

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

    # doc: end-embedding-pooling-params

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


555
class CompletionLogProbs(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
556
557
558
    text_offset: List[int] = Field(default_factory=list)
    token_logprobs: List[Optional[float]] = Field(default_factory=list)
    tokens: List[str] = Field(default_factory=list)
559
560
    top_logprobs: List[Optional[Dict[str,
                                     float]]] = Field(default_factory=list)
Zhuohan Li's avatar
Zhuohan Li committed
561
562


563
class CompletionResponseChoice(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
564
565
    index: int
    text: str
566
    logprobs: Optional[CompletionLogProbs] = None
567
568
    finish_reason: Optional[str] = None
    stop_reason: Optional[Union[int, str]] = Field(
569
570
571
572
573
574
        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"),
    )
575
    prompt_logprobs: Optional[List[Optional[Dict[int, Logprob]]]] = None
Zhuohan Li's avatar
Zhuohan Li committed
576
577


578
class CompletionResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
579
580
581
582
583
584
585
586
    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


587
class CompletionResponseStreamChoice(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
588
589
    index: int
    text: str
590
    logprobs: Optional[CompletionLogProbs] = None
591
592
    finish_reason: Optional[str] = None
    stop_reason: Optional[Union[int, str]] = Field(
593
594
595
596
597
598
        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
599
600


601
class CompletionStreamResponse(OpenAIBaseModel):
Zhuohan Li's avatar
Zhuohan Li committed
602
603
604
605
606
    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]
607
    usage: Optional[UsageInfo] = Field(default=None)
608
609


610
class EmbeddingResponseData(OpenAIBaseModel):
611
612
    index: int
    object: str = "embedding"
613
    embedding: Union[List[float], str]
614
615


616
class EmbeddingResponse(OpenAIBaseModel):
617
618
619
620
621
622
623
624
    id: str = Field(default_factory=lambda: f"cmpl-{random_uuid()}")
    object: str = "list"
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    data: List[EmbeddingResponseData]
    usage: UsageInfo


625
626
627
628
629
630
631
632
633
634
635
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


636
class ChatMessage(OpenAIBaseModel):
637
638
    role: str
    content: str
639
    tool_calls: List[ToolCall] = Field(default_factory=list)
640
641


642
643
644
645
646
647
648
649
650
651
652
653
654
655
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


656
class ChatCompletionResponseChoice(OpenAIBaseModel):
657
658
    index: int
    message: ChatMessage
659
    logprobs: Optional[ChatCompletionLogProbs] = None
660
    finish_reason: Optional[str] = None
661
    stop_reason: Optional[Union[int, str]] = None
662
663


664
class ChatCompletionResponse(OpenAIBaseModel):
665
    id: str = Field(default_factory=lambda: f"chatcmpl-{random_uuid()}")
666
    object: Literal["chat.completion"] = "chat.completion"
667
668
669
670
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[ChatCompletionResponseChoice]
    usage: UsageInfo
671
    prompt_logprobs: Optional[List[Optional[Dict[int, Logprob]]]] = None
672
673


674
class DeltaMessage(OpenAIBaseModel):
675
676
    role: Optional[str] = None
    content: Optional[str] = None
677
    tool_calls: List[ToolCall] = Field(default_factory=list)
678
679


680
class ChatCompletionResponseStreamChoice(OpenAIBaseModel):
681
682
    index: int
    delta: DeltaMessage
683
    logprobs: Optional[ChatCompletionLogProbs] = None
684
    finish_reason: Optional[str] = None
685
    stop_reason: Optional[Union[int, str]] = None
686
687


688
class ChatCompletionStreamResponse(OpenAIBaseModel):
689
    id: str = Field(default_factory=lambda: f"chatcmpl-{random_uuid()}")
690
    object: Literal["chat.completion.chunk"] = "chat.completion.chunk"
691
692
693
    created: int = Field(default_factory=lambda: int(time.time()))
    model: str
    choices: List[ChatCompletionResponseStreamChoice]
694
    usage: Optional[UsageInfo] = Field(default=None)
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715


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

716
    # The parameters of the request.
717
    body: Union[ChatCompletionRequest, EmbeddingRequest]
718
719


720
721
722
723
724
725
726
727
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.
728
    body: Optional[Union[ChatCompletionResponse, EmbeddingResponse]] = None
729
730


731
732
733
734
735
736
737
738
739
740
741
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

742
    response: Optional[BatchResponseData]
743
744
745
746

    # For requests that failed with a non-HTTP error, this will contain more
    # information on the cause of the failure.
    error: Optional[Any]
747
748


749
750
751
752
753
754
755
756
757
758
759
class TokenizeCompletionRequest(OpenAIBaseModel):
    model: str
    prompt: str

    add_special_tokens: bool = Field(default=True)


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

760
761
    add_generation_prompt: bool = Field(default=True)
    add_special_tokens: bool = Field(default=False)
762
763
764


TokenizeRequest = Union[TokenizeCompletionRequest, TokenizeChatRequest]
765
766
767
768
769


class TokenizeResponse(OpenAIBaseModel):
    count: int
    max_model_len: int
770
    tokens: List[int]
771
772
773
774
775
776
777
778
779


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


class DetokenizeResponse(OpenAIBaseModel):
    prompt: str