vllm.renderers.kimi ¶

class KimiRenderer(HfRenderer):
    """Renderer for Kimi models.

    Inherits from HfRenderer but uses KimiTokenizer instead of CachedHfTokenizer
    to properly handle TikTokenTokenizer-based models.
    """

    @classmethod
    def from_config(  # type: ignore[override]
        cls,
        config: VllmConfig,
        tokenizer_kwargs: dict[str, Any],
    ) -> "KimiRenderer":
        model_config = config.model_config
        if model_config.skip_tokenizer_init:
            tokenizer = None
        else:
            tokenizer = cast(
                KimiTokenizer,
                cached_get_tokenizer(
                    tokenizer_cls=KimiTokenizer,
                    **tokenizer_kwargs,
                ),
            )

        return cls(config, tokenizer)

    def __init__(
        self,
        config: VllmConfig,
        tokenizer: KimiTokenizer | None,
    ) -> None:
        super().__init__(config, tokenizer)

        self._kimia_prompt_prefix = "<|im_kimia_user_msg_start|>"
        self._kimia_prompt_suffix = "<|im_msg_end|><|im_kimia_assistant_msg_start|>"

    def get_tokenizer(self) -> KimiTokenizer:
        tokenizer = self.tokenizer
        if tokenizer is None:
            raise ValueError("Tokenizer not available when `skip_tokenizer_init=True`")
        return tokenizer

    def render_prompt(self, prompt):
        prompt = super().render_prompt(prompt)
        if (
            isinstance(prompt, dict)
            and "prompt" in prompt
            and "prompt_token_ids" not in prompt
        ):
            prompt_text = prompt["prompt"]
            if isinstance(prompt_text, str) and not self._is_kimia_wrapped(prompt_text):
                prompt["prompt"] = (
                    f"{self._kimia_prompt_prefix}{prompt_text}"
                    f"{self._kimia_prompt_suffix}"
                )
        return prompt

    def _is_kimia_wrapped(self, prompt_text: str) -> bool:
        return (
            self._kimia_prompt_prefix in prompt_text
            or "<|im_kimia_assistant_msg_start|>" in prompt_text
            or "<|im_msg_end|>" in prompt_text
        )