transformers-js

Transformers.js — Machine Learning for JavaScript

Runs state-of-the-art ML models directly in JavaScript, in browsers and server-side runtimes (Node.js, Bun, Deno), with no Python server required.

Installation

npm install @huggingface/transformers

// Browser (CDN)
import { pipeline } from 'https://cdn.jsdelivr.net/npm/@huggingface/transformers';

Core Concepts

Pipeline API — groups preprocessing, inference, and postprocessing. Always dispose() when done to free memory (see references/EXAMPLES.md for cleanup patterns):

import { pipeline } from '@huggingface/transformers';
const pipe = await pipeline('sentiment-analysis');
const result = await pipe('I love transformers!');
await pipe.dispose();

Model selection — pass a model ID as the second argument, e.g. pipeline('sentiment-analysis', 'Xenova/bert-base-multilingual-uncased-sentiment'). Browse compatible models at https://huggingface.co/models?library=transformers.js&sort=trending, filtered by pipeline_tag for a specific task.

Device: { device: 'webgpu' } for GPU acceleration (falls back to WASM/CPU when unsupported); omit for CPU/WASM default.

Quantization: { dtype: 'q4' } — options fp32 (largest/most accurate), fp16, q8, q4 (smallest, some accuracy loss).

Supported Tasks

One pipeline call per task, e.g. await pipeline('image-classification')('https://example.com/image.jpg'). Task IDs by category:

• NLP: text-classification/sentiment-analysis, token-classification/ner, question-answering, fill-mask, summarization, translation, text-generation, text2text-generation, zero-shot-classification
• Vision: image-classification, object-detection, image-segmentation, depth-estimation, zero-shot-image-classification, image-to-image
• Audio: automatic-speech-recognition, audio-classification, text-to-speech/text-to-audio
• Multimodal: image-to-text, document-question-answering, zero-shot-object-detection
• Embeddings: feature-extraction (add { pooling: 'mean', normalize: true } for sentence embeddings), sentence-similarity