Function gatherQmm

gatherQmm(
    x: MlxArray,
    w: MlxArray,
    scales: MlxArray,
    biases: MlxArray | null,
    rhsIndices: MlxArray,
    spec: QuantSpec,
    sortedIndices: boolean,
    transpose?: boolean,
    s?: bigint,
): MlxArray
Gathered quantized matmul over stacked expert weights (MoE). x [..., 1, K] with rhs_indices [...] selecting the expert per row; port of mx.gather_qmm as QuantizedSwitchLinear uses it.
Parameters
- x: MlxArray
- w: MlxArray
- scales: MlxArray
- biases: MlxArray | null
- rhsIndices: MlxArray
- spec: QuantSpec
- sortedIndices: boolean
- transpose: boolean = true
- s: bigint = gpuStream
Returns MlxArray
- Defined in packages/mlx/src/ops.ts:280