Language Model Networks: Supervision-Efficient Learning through Dense Communication

Language models are increasingly used not only as standalone predictors but also as components in larger inference systems, from test-time scaling to multi-agent collaboration. We study language model networks, where pre-trained language models serve as reusable nodes and intelligence emerges from their topology, communication, and optimization. Existing systems mostly communicate through natural language: easy to deploy, but discrete, inefficient, and hard to optimize from end-task supervision. We propose LMNet, a dense and differentiable realization of this paradigm. LMNet uses stripped LLMs as vertex modules and trainable seq2seq modules as communication edges, enabling intermediate nodes to exchange dense vectors while preserving natural-language input and output at the system boundary. By bypassing intermediate embedding and de-embedding, LMNet enables efficient information transfer, end-to-end gradient optimization, and learned communication beyond hand-designed protocols. Experiments show performance with small additional training cost and effective adaptation under limited supervision.

Modern AI language models are often used together: one model may draft an answer, another may check it, and another may combine the results. Today, these models usually communicate by writing text to each other, just as they communicate with people. This is easy to set up, but it can be slow, lossy, and difficult to improve automatically. We introduce LMNet, a way for language models inside the same system to share information more directly. Instead of forcing every intermediate message to be written as human-readable text, LMNet lets models pass richer internal signals to one another, while still taking normal text as input and producing normal text as output. The system can then learn how its parts should communicate by looking only at the final task result, without needing humans to label the intermediate steps. Our experiments show that this learned communication can improve reasoning and adaptation to new tasks, even when only limited training data is available. This suggests a path toward AI systems that reuse existing language models more effectively by teaching them how to work together.