Class AneTrellisMlpSplitK3i

Whole-MLP split for a 3-bit block-interleaved down projection.

Implements

Disposable

Index

Properties

channels down gate gpu program rest rows up

Methods

[dispose] dispose forward build

Properties

`Readonly`channels

channels: number

`Readonly`down

down: TrellisLinear

`Readonly`gate

gate: TrellisLinear

`Readonly`gpu

gpu: GpuRest

`Readonly`program

program: AneProgram

`Readonly`rest

rest: Rest

`Readonly`rows

rows: number

`Readonly`up

up: TrellisLinear

Methods

[dispose]

"[dispose]"(): void
Returns void
Implementation of Disposable.[dispose]
- Defined in packages/inference/src/layers/ane-prefill-split.ts:199

dispose

dispose(): void
As AneTrellisMlpSplit.dispose.

Returns void
- Defined in packages/inference/src/layers/ane-prefill-split.ts:201

forward

forward(hidden: MlxArray, swiglu: Swiglu): MlxArray
hidden [.., M, D], M <= rows → MLP(hidden) [.., M, D], unevaluated (mlpSplit).
Parameters
- hidden: MlxArray
- swiglu: Swiglu
Returns MlxArray
- Defined in packages/inference/src/layers/ane-prefill-split.ts:195

`Static`build

build(
    gate: TrellisLinear,
    up: TrellisLinear,
    down: TrellisLinear,
    fraction: number,
    rows: number,
    share?: AneMlpSplit,
    slices?: number,
    gpuRows?: AneGpuMlpRows,
): AneTrellisMlpSplitK3i
As AneTrellisMlpSplit.build.
Parameters
- gate: TrellisLinear
- up: TrellisLinear
- down: TrellisLinear
- fraction: number
- rows: number
- Optionalshare: AneMlpSplit
- Optionalslices: number
- gpuRows: AneGpuMlpRows = legacyGpuRows
Returns AneTrellisMlpSplitK3i
- Defined in packages/inference/src/layers/ane-prefill-split.ts:186