Low-rank tuning cuts AI chatbot reply length without losing quality

Original: LOCUS: Task-Aware Low-Rank Post-Training for Token-Efficient Language Generation

Source: arxiv.org ↗