README.md
3.1 KB · 82 lines · markdown Raw
1 ---
2 base_model: BAAI/bge-base-en-v1.5
3 library_name: transformers.js
4 license: mit
5 ---
6
7 https://huggingface.co/BAAI/bge-base-en-v1.5 with ONNX weights to be compatible with Transformers.js.
8
9 ## Usage (Transformers.js)
10
11 If you haven't already, you can install the [Transformers.js](https://huggingface.co/docs/transformers.js) JavaScript library from [NPM](https://www.npmjs.com/package/@huggingface/transformers) using:
12 ```bash
13 npm i @huggingface/transformers
14 ```
15
16 You can then use the model to compute embeddings, as follows:
17
18 ```js
19 import { pipeline } from '@huggingface/transformers';
20
21 // Create a feature-extraction pipeline
22 const extractor = await pipeline('feature-extraction', 'Xenova/bge-base-en-v1.5');
23
24 // Compute sentence embeddings
25 const texts = ['Hello world.', 'Example sentence.'];
26 const embeddings = await extractor(texts, { pooling: 'mean', normalize: true });
27 console.log(embeddings);
28 // Tensor {
29 // dims: [ 2, 768 ],
30 // type: 'float32',
31 // data: Float32Array(1536) [ 0.019079938530921936, 0.041718777269124985, ... ],
32 // size: 1536
33 // }
34
35 console.log(embeddings.tolist()); // Convert embeddings to a JavaScript list
36 // [
37 // [ 0.019079938530921936, 0.041718777269124985, 0.037672195583581924, ... ],
38 // [ 0.020936904475092888, 0.020080938935279846, -0.00787576474249363, ... ]
39 // ]
40 ```
41
42 You can also use the model for retrieval. For example:
43 ```js
44 import { pipeline, cos_sim } from '@huggingface/transformers';
45
46 // Create a feature-extraction pipeline
47 const extractor = await pipeline('feature-extraction', 'Xenova/bge-base-en-v1.5');
48
49 // List of documents you want to embed
50 const texts = [
51 'Hello world.',
52 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.',
53 'I love pandas so much!',
54 ];
55
56 // Compute sentence embeddings
57 const embeddings = await extractor(texts, { pooling: 'mean', normalize: true });
58
59 // Prepend recommended query instruction for retrieval.
60 const query_prefix = 'Represent this sentence for searching relevant passages: '
61 const query = query_prefix + 'What is a panda?';
62 const query_embeddings = await extractor(query, { pooling: 'mean', normalize: true });
63
64 // Sort by cosine similarity score
65 const scores = embeddings.tolist().map(
66 (embedding, i) => ({
67 id: i,
68 score: cos_sim(query_embeddings.data, embedding),
69 text: texts[i],
70 })
71 ).sort((a, b) => b.score - a.score);
72 console.log(scores);
73 // [
74 // { id: 1, score: 0.7787772374597298, text: 'The giant panda (Ailuropoda melanoleuca), sometimes called a panda bear or simply panda, is a bear species endemic to China.' },
75 // { id: 2, score: 0.7071589521880506, text: 'I love pandas so much!' },
76 // { id: 0, score: 0.4252782730390429, text: 'Hello world.' }
77 // ]
78 ```
79
80 ---
81
82 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](https://huggingface.co/docs/optimum/index) and structuring your repo like this one (with ONNX weights located in a subfolder named `onnx`).