Interface DenoisingTraining

The denoising objective of a method: "denoising" graph.

interface DenoisingTraining {
    layerModulePaths: readonly string[];
    vocabSize: number;
    canvasLogits(promptIds: number[], canvas: MlxArray): MlxArray;
}

Index

Properties

layerModulePaths vocabSize

Methods

Properties

`Readonly`layerModulePaths

layerModulePaths: readonly string[]

LoRA module path of each decoder layer (<path>.self_attn.q_proj, ...).

`Readonly`vocabSize

vocabSize: number

Methods

canvasLogits

canvasLogits(promptIds: number[], canvas: MlxArray): MlxArray
Logits [1, L, vocab] (float32) of denoising canvas [1, L] given the prompt.
Parameters
- promptIds: number[]
- canvas: MlxArray
Returns MlxArray
- Defined in packages/inference/src/contracts/mlx/trainable.ts:90