mlx-bun library API
    Preparing search index...

    The same recurrence and projections with a fused normalization/gate tail.

    Type Parameters

    Hierarchy (View Summary)

    Index
    aLog: MlxArray
    convKernel: number
    convWeight: MlxArray
    dtBias: MlxArray
    eps: number
    headKDim: number
    headVDim: number
    inProjA: L
    inProjB: L
    inProjQkv: L
    inProjZ: L
    keyDim: number
    normWeight: MlxArray
    numKHeads: number
    numVHeads: number
    outProj: L
    qkScale: { k: MlxArray; q: MlxArray } | null = null

    Per-model seam; null retains the oracle graph (weightless rms_norm, then a scalar multiply). When set, the scale rides in as the norm's weight: MLX's kernel writes w * T(x * inv), the same bf16 product, in one kernel. bf16 only.

    What this block lends its recurrent cache on every call: its recurrence weights and the glue between the convolution and the recurrence.

    valueDim: number