Agents tagged "model-fusion"

Cache-to-Cache

Let independent LLMs exchange semantics directly through projected KV caches instead of generated text.