GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection

2 points| mau | 1 year ago |arxiv.org