text2vec

Why we included this project

Teams doing Chinese semantic search or duplicate detection often need to turn sentences and paragraphs into dense vectors, and text2vec makes that work straightforward. It packages Word2Vec, Sentence-BERT, and CoSENT behind one Python API, so you can try a fast literal-matching baseline and a heavier learned encoder without building each one from scratch. The project reports that CoSENT converges faster and beats classic SBERT on Chinese matching benchmarks, and it publishes per-model QPS and STS results so you can pick a sensible default before committing. Pretrained Chinese models ship ready to use, and the training code is there if you want to fine-tune on your own data. For teams working mostly in Chinese, that combination is usually the difference between a weekend prototype and wiring a sentence encoder into production yourself.