Class CreateCompletionRequest

examples/server/server.py:2504–2564

class CreateCompletionRequest(BaseModel):
    model_config = ConfigDict(extra="ignore")

    prompt: CompletionPrompt = ""
    suffix: Optional[str] = None
    max_tokens: Optional[int] = Field(default=16, ge=0)
    temperature: float = 0.8
    top_p: float = Field(default=0.95, ge=0.0, le=1.0)
    echo: bool = False
    stop: Optional[Union[str, List[str]]] = None
    stream: bool = False
    logprobs: Optional[int] = Field(default=None, ge=0)
    presence_penalty: Optional[float] = Field(default=0.0, ge=-2.0, le=2.0)
    frequency_penalty: Optional[float] = Field(default=0.0, ge=-2.0, le=2.0)
    logit_bias: Optional[Dict[str, float]] = None
    seed: Optional[int] = None
    model: Optional[str] = None
    n: int = Field(default=1, ge=1)
    best_of: Optional[int] = Field(default=None, ge=1)
    user: Optional[str] = None

    @field_validator("logit_bias")
    @classmethod
    def validate_logit_bias(cls, value: Optional[Dict[str, float]]) -> Optional[Dict[str, float]]:
        if value is None:
            return None
        result: Dict[str, float] = {}
        for key, bias in value.items():
            int(key)
            result[key] = float(bias)
        return result

    @model_validator(mode="after")
    def validate_after(self) -> "CreateCompletionRequest":
        if self.best_of is None:
            self.best_of = self.n
        if self.best_of < self.n:
            raise ValueError("best_of must be greater than or equal to n")
        if self.stream and self.best_of > 1:
            raise ValueError("best_of is not supported for streaming completions")
        if len(self.normalized_prompt()) > 1 and self.stream:
            raise ValueError("streaming does not support multiple prompts")
        return self

    def normalized_prompt(self) -> List[Union[str, List[int]]]:
        if isinstance(self.prompt, str):
            return [self.prompt]
        if all(isinstance(token, int) for token in self.prompt):
            return [cast(List[int], self.prompt)]
        if all(isinstance(prompt, str) for prompt in self.prompt):
            return cast(List[Union[str, List[int]]], list(cast(List[str], self.prompt)))
        if all(
            isinstance(prompt, list)
            and all(isinstance(token, int) for token in prompt)
            for prompt in self.prompt
        ):
            return cast(
                List[Union[str, List[int]]],