Embeddings

Embed text on device with ML Kit’s embedding model. It’s an alpha API, and not every device has a model.

import { android } from '@react-native-ai/android'
import { cosineSimilarity, embedMany } from 'ai'

const model = android.embeddingModel({ taskType: 'semanticSimilarity' })

const { embeddings } = await embedMany({
  model,
  values: ['The cat sleeps on the sofa.', 'A kitten naps on the couch.'],
})

console.log(cosineSimilarity(embeddings[0], embeddings[1]))

taskType tells the model what the embeddings are for: retrievalQuery, retrievalDocument, semanticSimilarity, classification, clustering, questionAnswering, factVerification or codeRetrieval.

Availability and Download

const info = await model.getModelInfo()
// { availability: 'available', dimension: 768, modelSignature: '...' }

await model.prepare({ onProgress: (progress) => console.log(progress) })

availability is unavailable when the device has no embedding model. Embedding prepares the model on its own when you don’t, and logs a warning.