GaLore proposes a memory-efficient training strategy for large language models, reducing memory usage while maintaining performance.
