mlx-bun library API
    Preparing search index...
    interface Qwen3VLPreprocessed {
        cols: number;
        gridThw: [number, number, number];
        groupTimestamps?: number[];
        imageTokens: number;
        pixelValues: Float32Array;
        rows: number;
    }
    Index
    cols: number
    gridThw: [number, number, number]

    [t, h, w] in PATCH units (h = resizedH/16 etc.).

    groupTimestamps?: number[]

    Videos only: per-temporal-group timestamps in seconds (frame-pair averaged, transformers _calculate_timestamps) — the <X.X seconds> headers the prompt builder renders before each frame group.

    imageTokens: number

    Language-side image token count = t * (h/merge) * (w/merge).

    pixelValues: Float32Array

    [gridTgridHgridW, Ctpsps*ps] f32 patch rows (merge-block-major).

    rows: number