Class Completion

examples/server/server.py:3807–3846

@dataclass
class Completion:
    request_id: str
    index: int
    seq_id: int
    sampler: "Sampler"
    prompt_tokens: List[int]
    prompt_length: int
    prompt_text: str
    multimodal_prompt: bool
    max_total_tokens: int
    stop_sequences: List[bytes]
    logprobs: Optional[int]
    completion_tokens: List[int] = field(default_factory=list)
    token_records: List[Token] = field(default_factory=list)
    rendered_bytes: bytearray = field(default_factory=bytearray)
    detokenized_prefix_bytes: bytearray = field(default_factory=bytearray)
    pending_input_tokens: List[int] = field(default_factory=list)
    draft_tokens: List[int] = field(default_factory=list)
    pending_finish_reason: Optional[str] = None
    returned_token_count: int = 0
    finished: bool = False
    finish_reason: Optional[str] = None
    score_sum: float = 0.0
    rank_by_score: bool = False

    @property
    def total_tokens(self) -> int:
        return self.prompt_length + len(self.completion_tokens)

    @property
    def completion_token_count(self) -> int:
        return len(self.completion_tokens)

    @property
    def needs_token_logprob(self) -> bool:
        return self.logprobs is not None or self.rank_by_score

    @property
    def max_stop_sequence_length(self) -> int:
        return max((len(stop) for stop in self.stop_sequences), default=0)


@dataclass