https://huggingface.co/BAAI/bge-large-en-v1.5 with ONNX weights to be compatible with Transformers.js.
Usage (Transformers.js)
If you haven't already, you can install the Transformers.js JavaScript library from NPM using:
npm i @huggingface/transformers
You can then use the model to compute embeddings, as follows:
import { pipeline } from '@huggingface/transformers';
// Create a feature-extraction pipeline
const extractor = await pipeline('feature-extraction', 'Xenova/bge-large-en-v1.5');
// Compute sentence embeddings
const texts = [ 'Hello world.', 'Example sentence.'];
const embeddings = await extractor(texts, { pooling: 'mean', normalize: true });
console.log(embeddings);
// Tensor {
// dims: [ 2, 1024 ],
// type: 'float32',
// data: Float32Array(2048) [ 0.03169844672083855, 0.011085662990808487, ... ],
// size: 2048
// }
console.log(embeddings.tolist()); // Convert embeddings to a JavaScript list
// [
// [ 0.03169844672083855, 0.011085662990808487, 0.030054178088903427, ... ],
// [ 0.009418969973921776, -0.024539148434996605, 0.036459196358919144, ... ]
// ]
You can also use the model for retrieval. For example:
import { pipeline, cos_sim } from '@huggingface/transformers';
// Create a feature-extraction pipeline
const extractor = await pipeline('feature-extraction', 'Xenova/bge-large-en-v1.5');
// List of documents you want to embed
const texts = [
'Hello world.',
'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.',
'I love pandas so much!',
];
// Compute sentence embeddings
const embeddings = await extractor(texts, { pooling: 'mean', normalize: true });
// Prepend recommended query instruction for retrieval.
const query_prefix = 'Represent this sentence for searching relevant passages: '
const query = query_prefix + 'What is a panda?';
const query_embeddings = await extractor(query, { pooling: 'mean', normalize: true });
// Sort by cosine similarity score
const scores = embeddings.tolist().map(
(embedding, i) => ({
id: i,
score: cos_sim(query_embeddings.data, embedding),
text: texts[i],
})
).sort((a, b) => b.score - a.score);
console.log(scores);
// [
// { id: 1, score: 0.7671812872502833, text: 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.' },
// { id: 2, score: 0.7219157959783322, text: 'I love pandas so much!' },
// { id: 0, score: 0.5109676329796601, text: 'Hello world.' }
// ]
Note: Having a separate repo for ONNX weights is intended to be a temporary solution until WebML gains more traction. If you would like to make your models web-ready, we recommend converting to ONNX using 🤗 Optimum and structuring your repo like this one (with ONNX weights located in a subfolder named onnx).