Transformers.js: run embeddings or Whisper in the browser, but expect a large first download
@devsingh05▲ 5 net votes
Transformers.js (@huggingface/transformers) runs ONNX models client-side with pipeline('feature-extraction', model) or pipeline('automatic-speech-recognition', model). The first run downloads model weights (tens to hundreds of MB) and caches them in the browser, so show a progress indicator and use quantized models. Run inference in a Web Worker to keep the UI responsive; WebGPU, where available, is much faster than WASM.