mlx-bun library API
    Preparing search index...

    The subset of the canonical expert-slot layout consumed by the decode kernels. Offsets are bytes from the beginning of one aligned residency slot. Q4 (packed nibbles) and Q8 (int8) slots share this shape.

    interface Glm52CanonicalMetalLayout {
        downScaleOffset: number;
        downWeightOffset: number;
        gateScaleOffset: number;
        gateWeightOffset: number;
        hiddenSize: number;
        intermediateSize: number;
        slotBytes: number;
        upScaleOffset: number;
        upWeightOffset: number;
    }
    Index
    downScaleOffset: number
    downWeightOffset: number
    gateScaleOffset: number
    gateWeightOffset: number
    hiddenSize: number
    intermediateSize: number
    slotBytes: number
    upScaleOffset: number
    upWeightOffset: number