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

4 points | by ibobev 4 hours ago