IntuitionEngineeringTeam/chars2vec

Uses LSTM layers to process character sequences, generating fixed-dimensional embeddings on-the-fly without storing a lookup dictionary—enabling robust handling of misspellings, slang, and out-of-vocabulary words. Trainable on custom datasets via contrastive learning with word pair similarity labels, with five pretrained English models (50-300 dimensions) available. Supports Python 2.7 and 3.0+ with a simple API for vectorizing words and training custom models.

174 stars. No commits in the last 6 months. Available on PyPI.