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

3 points | by pretext 12 hours ago