More efficient word embeddings for production classification tasks, with smaller models trained on supervised data & reduced-size conventional pre-trained embeddings.
Based on :
- FB's SentEval for multi-task supervision
- HuggingFace's Transformers for access to pre-trained models
- SemEval tasks for transfer evaluation