How to use sentence-transformers/static-similarity-mrl-multilingual-v1 model?

[michalkvasnicak]

Question

If I try to use sentence-transformers/static-similarity-mrl-multilingual-v1 it fails on tokenizer.json not found. Is it possible to somehow convert the model to use it ? ONNX runtime is already there.

[xenova]

Sure, you can do the following:

import { AutoModel, AutoTokenizer, matmul } from '@huggingface/transformers';

// Load model and tokenizer
const model = await AutoModel.from_pretrained('sentence-transformers/static-similarity-mrl-multilingual-v1', {
    config: { model_type: 'bert' },
});
const tokenizer = await AutoTokenizer.from_pretrained('Xenova/bert-base-multilingual-uncased');

const sentences = [
    'It is known for its dry red chili powder .',
    'It is popular for dry red chili powder .',
    'These monsters will move in large groups .',
];
const inputs = tokenizer(sentences, { padding: true, truncation: true });
const { sentence_embedding } = await model(inputs);

const normalized = sentence_embedding.normalize();
const scores = await matmul(normalized, normalized.transpose(1, 0));
console.log(scores.tolist());
// [
//   [ 0.9999996423721313, 0.8800840973854065, 0.0017940629040822387 ],
//   [ 0.8800840973854065, 1.0000007152557373, 0.03064093366265297 ],
//   [ 0.0017940629040822387, 0.03064093366265297, 1.0000004768371582 ]
// ]

There are some warnings printed out when loading the model, but you can safely ignore those.