Tag
The paper introduces Latent Cache Flow (LCF), a method for efficient model-to-model communication by exchanging compressed KV caches instead of text, reducing adapter size and enabling cross-context communication.