Cache-to-Cache: Direct Semantic Communication Between LLMs (2025)

A new paradigm called Cache-to-Cache (C2C) enables direct semantic communication between Large Language Models (LLMs), addressing limitations of text-based communication. C2C projects and fuses the KV-cache of source and target models using a neural network, allowing direct semantic transfer and avoiding explicit intermediate text generation. Experiments show C2C achieves 6.4-14.2% higher average accuracy than individual models and outperforms text communication by 3.1-5.4%, with a 2.5x speedup in latency. This method leverages rich semantic information for improved performance and efficiency.

This paper introduces Cache-to-Cache, a novel paradigm for direct LLM semantic communication, unlike previous methods that rely on text, achieving 2.5x speedup and higher accuracy.