vllm/cacheflow/entrypoints/openai/protocol.py

# Adapted from https://github.com/lm-sys/FastChat/blob/168ccc29d3f7edc50823016105c024fe2282732a/fastchat/protocol/openai_api_protocol.py
import time
from typing import Dict, List, Literal, Optional, Union

from pydantic import BaseModel, Field

from cacheflow.utils import random_uuid


class ErrorResponse(BaseModel):
    object: str = "error"
    message: str
    type: str
    param: Optional[str] = None
    code: Optional[str] = None


class ModelPermission(BaseModel):
    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
    is_blocking: str = False


class ModelCard(BaseModel):
    id: str
    object: str = "model"
    created: int = Field(default_factory=lambda: int(time.time()))
    owned_by: str = "cacheflow"
    root: Optional[str] = None
    parent: Optional[str] = None
    permission: List[ModelPermission] = Field(default_factory=list)


class ModelList(BaseModel):
    object: str = "list"
    data: List[ModelCard] = Field(default_factory=list)


class UsageInfo(BaseModel):
    prompt_tokens: int = 0
    total_tokens: int = 0
    completion_tokens: Optional[int] = 0


class ChatCompletionRequest(BaseModel):
    model: str
    messages: List[Dict[str, str]]
    temperature: Optional[float] = 0.7
    top_p: Optional[float] = 1.0
    n: Optional[int] = 1
    max_tokens: Optional[int] = None
    stop: Optional[Union[str, List[str]]] = None
    stream: Optional[bool] = False
    presence_penalty: Optional[float] = 0.0
    frequency_penalty: Optional[float] = 0.0
    user: Optional[str] = None


class CompletionRequest(BaseModel):
    model: str
    prompt: str
    suffix: Optional[str] = None
    max_tokens: Optional[int] = 16
    temperature: Optional[float] = 1.0
    top_p: Optional[float] = 1.0
    n: Optional[int] = 1
    stream: Optional[bool] = False
    logprobs: Optional[int] = None
    echo: Optional[bool] = False
    stop: Optional[Union[str, List[str]]] = Field(default_factory=list)
    presence_penalty: Optional[float] = 0.0
    frequency_penalty: Optional[float] = 0.0
    best_of: Optional[int] = None
    logit_bias: Optional[Dict[str, float]] = None
    user: Optional[str] = None
    # Additional parameters supported by cacheflow
    top_k: Optional[int] = -1
    ignore_eos: Optional[bool] = False
    use_beam_search: Optional[bool] = False


class LogProbs(BaseModel):
    text_offset: List[int] = Field(default_factory=list)
    token_logprobs: List[Optional[float]] = Field(default_factory=list)
    tokens: List[str] = Field(default_factory=list)
    top_logprobs: List[Optional[Dict[str, float]]] = Field(default_factory=list)


class CompletionResponseChoice(BaseModel):
    index: int
    text: str
    logprobs: Optional[LogProbs] = None
    finish_reason: Optional[Literal["stop", "length"]] = None


class CompletionResponse(BaseModel):
    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


class CompletionResponseStreamChoice(BaseModel):
    index: int
    text: str
    logprobs: Optional[LogProbs] = None
    finish_reason: Optional[Literal["stop", "length"]] = None


class CompletionStreamResponse(BaseModel):
    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]