Class QuantizedEmbedding

constructor

• new QuantizedEmbedding(
      w: MlxArray,
      scales: MlxArray,
      biases: MlxArray | null,
      spec: QuantSpec,
  ): QuantizedEmbedding

  Parameters

  • w: MlxArray
  • scales: MlxArray
  • biases: MlxArray | null
  • spec: QuantSpec

  Returns QuantizedEmbedding

Readonlybiases

Readonlyscales

Readonlyspec

Readonlyw

asLinear

• asLinear(h: MlxArray): MlxArray

  Tied output head: h [1, L, hidden] → logits [1, L, vocab].

  Parameters

  • h: MlxArray

  Returns MlxArray

encode

• encode(ids: MlxArray): MlxArray

  ids [1, L] (int/uint) → embeddings [1, L, hidden].

  Parameters

  • ids: MlxArray

  Returns MlxArray

Staticload

• load(weights: Weights, path: string, config: ModelConfig): QuantizedEmbedding

  Parameters

  • weights: Weights
  • path: string
  • config: ModelConfig

  Returns QuantizedEmbedding