Full Professor, Beijing Institute of Technology
2 papers at NeurIPS 2025
We propose Fira, a plug-and-play memory-efficient training framework of LLMs to enable full-rank training consistently under the low-rank constraint.
