Transformers.js

Use it when

• You want text classification, embeddings, translation, object detection, or speech recognition inside a web app without standing up any inference backend
• User audio, images, or documents must stay on-device, so inference has to run client-side in the browser via WASM or WebGPU

What it solves

Not the fit when

• Unsupported tasks in the README matrix, including text-to-image, visual question answering, mask generation, video classification, and tabular tasks
• High-throughput backend serving; the primary surface is JavaScript inference in browsers and compatible JavaScript runtimes
• Models must have compatible ONNX artifacts, or be converted with Optimum
• The default configuration downloads models from Hugging Face and WASM binaries from a CDN; set localModelPath, allowRemoteModels=false, and a local wasmPaths value when offline asset loading is required
• The Apache-2.0 library license does not replace the separate license and usage terms of each downloaded model
• training or fine-tuning models
• high-throughput backend model serving
• text-to-image generation

Install

npm i @huggingface/transformers (or import from the jsDelivr CDN as an ES module)

Invoke

import { pipeline } from '@huggingface/transformers'; const pipe = await pipeline('sentiment-analysis'); await pipe('I love transformers!') - pass { device: 'webgpu' } or { dtype: 'q4' } for GPU or quantized runs

Alternatives

No reviewed alternatives recorded yet.