vllm.entrypoints.pooling.scoring.typing ¶

class ScoreMultiModalParam(TypedDict, total=False):
    """
    A specialized parameter type for scoring multimodal content

    The reasons why don't reuse `CustomChatCompletionMessageParam` directly:
    1. Score tasks don't need the 'role' field (user/assistant/system) that's required in chat completions
    2. Including chat-specific fields would confuse users about their purpose in scoring
    3. This is a more focused interface that only exposes what's needed for scoring
    """  # noqa: E501

    content: Required[list[ScoreContentPartParam]]
    """The multimodal contents"""