Alibaba-NLP (mGTE) представляет собой мощную мультиязычную модель для семантического поиска и реранкинга текстов. Главная фишка этого ИИ — поддержка длинного контекста до 8192 токенов, что позволяет эффективно обрабатывать объемные документы на множестве языков.
We present systematic efforts in building longcontext multilingual text representation model (TRM) and reranker from scratch for text retrieval. We first introduce a text encoder (base size) enhanced with RoPE and unpadding, pretrained in a native 8192-token context (longer than 512 of previous multilingual encoders). Then we construct a hybrid TRM and a crossencoder reranker by contrastive learning. Evaluations show that our text encoder outperforms the same-sized previous state-of-the-art XLMR. Meanwhile, our TRM and reranker match the performance of large-sized state-of-the-art BGE-M3 models and achieve better results on long-context retrieval benchmarks. Further analysis demonstrate that our proposed models exhibit higher efficiency during both training and inference. We believe their efficiency and effectiveness could benefit various researches and industrial applications.