Recent Developments in LLM Architectures: KV Sharing, MHC, Compressed Attention

2 points | by eigenBasis 9 hours ago