bilingual-document-embedding
bilingual-document-embedding is the Embedding Model for document in bilingual language: french and english with context length up to 8096 tokens . This model is a specialized sentence-embedding trained specifically for the bilingual language, leveraging the robust capabilities of BGE M3, a pre-trained language model larged on the BGE M3 architecture. The model utilizes xlm-roberta to encode english-french sentences into a 1024-dimensional vector space, facilitating a wide range of applications from semantic search to text clustering. The embeddings capture the nuanced meanings of english-french sentences, reflecting both the lexical and contextual layers of the language.
Full Model Architecture
SentenceTransformer(
(0): Transformer({'max_seq_length': 512, 'do_lower_case': False}) with Transformer model: BilingualModel
(1): Pooling({'word_embedding_dimension': 1024, 'pooling_mode_cls_token': False, 'pooling_mode_mean_tokens': True, 'pooling_mode_max_tokens': False, 'pooling_mode_mean_sqrt_len_tokens': False, 'pooling_mode_weightedmean_tokens': False, 'pooling_mode_lasttoken': False, 'include_prompt': True})
(2): Normalize()
)
Training and Fine-tuning process
Stage 1: NLI Training
- Dataset: [(SNLI+XNLI) for english+french]
- Method: Training using Multi-Negative Ranking Loss. This stage focused on improving the model's ability to discern and rank nuanced differences in sentence semantics.
Stage 3: Continued Fine-tuning for Semantic Textual Similarity on STS Benchmark
- Dataset: [STSB-fr and en]
- Method: Fine-tuning specifically for the semantic textual similarity benchmark using Siamese BERT-Networks configured with the 'sentence-transformers' library.
Stage 4: Advanced Augmentation Fine-tuning
- Dataset: STSB with generate silver sample from gold sample
- Method: Employed an advanced strategy using Augmented SBERT with Pair Sampling Strategies, integrating both Cross-Encoder and Bi-Encoder models. This stage further refined the embeddings by enriching the training data dynamically, enhancing the model's robustness and accuracy.
Usage:
Using this model becomes easy when you have sentence-transformers installed:
pip install -U sentence-transformers
Then you can use the model like this:
from sentence_transformers import SentenceTransformer
sentences = ["Paris est une capitale de la France", "Paris is a capital of France"]
model = SentenceTransformer('Lajavaness/bilingual-document-embedding', trust_remote_code=True)
print(embeddings)
Evaluation
TODO
Citation
@article{chen2024bge,
title={Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation},
author={Chen, Jianlv and Xiao, Shitao and Zhang, Peitian and Luo, Kun and Lian, Defu and Liu, Zheng},
journal={arXiv preprint arXiv:2402.03216},
year={2024}
}
@article{conneau2019unsupervised,
title={Unsupervised cross-lingual representation learning at scale},
author={Conneau, Alexis and Khandelwal, Kartikay and Goyal, Naman and Chaudhary, Vishrav and Wenzek, Guillaume and Guzm{\'a}n, Francisco and Grave, Edouard and Ott, Myle and Zettlemoyer, Luke and Stoyanov, Veselin},
journal={arXiv preprint arXiv:1911.02116},
year={2019}
}
@article{reimers2019sentence,
title={Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks},
author={Nils Reimers, Iryna Gurevych},
journal={https://arxiv.org/abs/1908.10084},
year={2019}
}
@article{thakur2020augmented,
title={Augmented SBERT: Data Augmentation Method for Improving Bi-Encoders for Pairwise Sentence Scoring Tasks},
author={Thakur, Nandan and Reimers, Nils and Daxenberger, Johannes and Gurevych, Iryna},
journal={arXiv e-prints},
pages={arXiv--2010},
year={2020}