LEALLA

Usage

Using the model:

import torch
from transformers import BertModel, BertTokenizerFast

tokenizer = BertTokenizerFast.from_pretrained("setu4993/LEALLA-base")
model = BertModel.from_pretrained("setu4993/LEALLA-base")
model = model.eval()

english_sentences = [
    "dog",
    "Puppies are nice.",
    "I enjoy taking long walks along the beach with my dog.",
]
english_inputs = tokenizer(english_sentences, return_tensors="pt", padding=True)

with torch.no_grad():
    english_outputs = model(**english_inputs)

To get the sentence embeddings, use the pooler output:

english_embeddings = english_outputs.pooler_output

Output for other languages:

italian_sentences = [
    "cane",
    "I cuccioli sono carini.",
    "Mi piace fare lunghe passeggiate lungo la spiaggia con il mio cane.",
]
japanese_sentences = ["犬", "子犬はいいです", "私は犬と一緒にビーチを散歩するのが好きです"]
italian_inputs = tokenizer(italian_sentences, return_tensors="pt", padding=True)
japanese_inputs = tokenizer(japanese_sentences, return_tensors="pt", padding=True)

with torch.no_grad():
    italian_outputs = model(**italian_inputs)
    japanese_outputs = model(**japanese_inputs)

italian_embeddings = italian_outputs.pooler_output
japanese_embeddings = japanese_outputs.pooler_output

For similarity between sentences, an L2-norm is recommended before calculating the similarity:

import torch.nn.functional as F

def similarity(embeddings_1, embeddings_2):
    normalized_embeddings_1 = F.normalize(embeddings_1, p=2)
    normalized_embeddings_2 = F.normalize(embeddings_2, p=2)
    return torch.matmul(
        normalized_embeddings_1, normalized_embeddings_2.transpose(0, 1)
    )

print(similarity(english_embeddings, italian_embeddings))
print(similarity(english_embeddings, japanese_embeddings))
print(similarity(italian_embeddings, japanese_embeddings))

Details

Details about data, training, evaluation and performance metrics are available in the original paper.

BibTeX entry and citation info

@inproceedings{mao-nakagawa-2023-lealla,
    title = "{LEALLA}: Learning Lightweight Language-agnostic Sentence Embeddings with Knowledge Distillation",
    author = "Mao, Zhuoyuan  and
      Nakagawa, Tetsuji",
    booktitle = "Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics",
    month = may,
    year = "2023",
    address = "Dubrovnik, Croatia",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.eacl-main.138",
    doi = "10.18653/v1/2023.eacl-main.138",
    pages = "1886--1894",
    abstract = "Large-scale language-agnostic sentence embedding models such as LaBSE (Feng et al., 2022) obtain state-of-the-art performance for parallel sentence alignment. However, these large-scale models can suffer from inference speed and computation overhead. This study systematically explores learning language-agnostic sentence embeddings with lightweight models. We demonstrate that a thin-deep encoder can construct robust low-dimensional sentence embeddings for 109 languages. With our proposed distillation methods, we achieve further improvements by incorporating knowledge from a teacher model. Empirical results on Tatoeba, United Nations, and BUCC show the effectiveness of our lightweight models. We release our lightweight language-agnostic sentence embedding models LEALLA on TensorFlow Hub.",
}