mlx-bun library API
    Preparing search index...

    Interface MultimodalPrompt

    interface MultimodalPrompt {
        bidirMask: MlxArray | null;
        embeddings: MlxArray;
        ids: number[];
        multimodalMask: MlxArray;
        prefixIdentity?: string;
    }
    Index
    bidirMask embeddings ids multimodalMask prefixIdentity?
    bidirMask: MlxArray | null

    bool [L] image-token mask for the bidirectional attention overlay. null when the prompt contains ANY audio (§3.3 Q1: mixed image+audio runs fully causal) or no images. Caller disposes when non-null.

    embeddings: MlxArray

    Unscaled merged embeddings [1, L, hidden] — caller disposes.

    ids: number[]
    multimodalMask: MlxArray

    bool [L] union multimodal soft-token mask (image | audio) for per-layer-input id zeroing (§3.3 Q2). Caller disposes.

    prefixIdentity?: string

    Includes every media item and the exact token prefix through the last one.

    Settings
    Member Visibility

    On This Page
    Properties