Research graph
References from Characterizing KV-Cache Performance in Large Language Model Inference: Context Length, Attention Architecture, and GPU Memory Residency. Local targets link to admitted publications; unresolved targets remain external evidence.
