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