3 ms·Deep Optimizer States: Scalable Training of Transformer Interleaved Offloading1 points by sandwichsphinx 2y ago