@anvia/transformers API reference

Import all public symbols from @anvia/transformers.

import {
  adaptTransformersEmbeddingModel,
  DEFAULT_TRANSFORMERS_EMBEDDING_MODEL,
  loadTransformersEmbeddingModel,
  type AdaptTransformersEmbeddingModelOptions,
  type LoadedTransformersEmbeddingModel,
  type LoadTransformersEmbeddingModelOptions,
  type TransformersEmbeddingModelHandle,
  type TransformersFeatureExtractionPipeline,
  type TransformersPooling,
  type TransformersTensor,
} from '@anvia/transformers'

Load a model

function loadTransformersEmbeddingModel(
  options: LoadTransformersEmbeddingModelOptions,
): Promise<LoadedTransformersEmbeddingModel>

type LoadTransformersEmbeddingModelOptions = {
  modelId: string
  pooling?: 'mean' | 'cls'
  normalize?: boolean
  maxBatchSize?: number
  device?: string | Record<string, string>
  dtype?: string | Record<string, string>
  cacheDir?: string
  localFilesOnly?: boolean
  revision?: string
}

LoadedTransformersEmbeddingModel implements EmbeddingModel and AsyncDisposable; call close() or use await using to release the owned pipeline.

Adapt an existing pipeline

function adaptTransformersEmbeddingModel(
  options: AdaptTransformersEmbeddingModelOptions,
): TransformersEmbeddingModelHandle

type AdaptTransformersEmbeddingModelOptions = {
  runtime: TransformersFeatureExtractionPipeline
  modelId: string
  pooling?: 'mean' | 'cls'
  normalize?: boolean
  maxBatchSize?: number
}

An adapted model does not own or dispose the supplied runtime.

Runtime contracts

type TransformersTensor = {
  tolist(): unknown
  dispose(): void
}

type TransformersFeatureExtractionPipeline = {
  (
    texts: string[],
    options: { pooling: TransformersPooling; normalize: boolean },
  ): Promise<TransformersTensor>
  dispose(): Promise<void>
}

const DEFAULT_TRANSFORMERS_EMBEDDING_MODEL = 'Xenova/all-MiniLM-L6-v2'