oborchers/Fast_Sentence_Embeddings

Implements three lightweight algorithms (Average, SIF, and uSIF) that aggregate pre-trained word embeddings into sentence vectors using Cython-optimized routines, achieving 300k-500k sentences/second on CPU. Integrates seamlessly with Gensim's Word2Vec and FastText models, supports disk-streaming and RAM-to-disk training for massive corpora, and provides hub access to pre-trained embeddings including GloVe, Word2Vec, and FastText variants.

625 stars. No commits in the last 6 months.