One Adapter Pair per Model: A Universal Activation Interface for Language Models
2026-08-10 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors introduce the Universal Activation Bus, a way to connect multiple language models by creating a shared space for their internal activations. Instead of building separate tools for each model, they train adapters that let different models use the same analysis and interpretation tools. This shared setup works across five tested models, allowing one model's parts to help another make predictions without extra training. Their work shows it’s possible to reuse tools across similar language models by agreeing on a common activation format.
activation-based toolslanguage modelshidden spacelinear encoder-decoderadaptersemantic neighborhoodprobesparse autoencodernatural-language interpretertransfer learning
Authors
Su-Hyeon Kim, Jiwan Mun, Yo-Sub Han
Abstract
Activation-based tools are usually tied to one model's native hidden space, requiring probes, sparse autoencoders, and natural-language interpreters to be rebuilt or rediscovered for each new language model. We present a Universal Activation Bus, a framework that provides a common activation interface across compatible language models. Using a small set of source models, we learn a shared dense space together with one lightweight linear encoder--decoder adapter pair per model. After source training, the interface is frozen; a new model joins by fitting only its adapter pair on unlabeled matched text. The resulting interface allows activation-based tools to be shared across connected models, including common probes and SAE features as well as access to an NLA originally trained for a different model. Across five models, semantically related texts form consistent neighborhoods in the shared space, and an onboarded model reuses these tools effectively without retraining them. We further show that an intermediate activation from one model can be used by another model's frozen upper layers to produce predictions. These results establish a stable, model-wise activation contract for reusable tools across compatible language models.