2 ms·Recent Developments in LLM Architectures: KV Sharing, MHC, Compressed Attention2 points by eigenBasis 5mo ago