Researchers Introduce Tiered Language Models For Differential LLM Access

@gspandana love this, keep going.

Can we allow multiple access levels within 1 model? We introduce TLMs, packing different memories🧠& capabilities🛠️ in different configurations of the same weights! Check our preprint https://arxiv.org/pdf/2606.21638! Lucky to have supervised the talented @arkil_patel & Charbel El Feghali

Can you release a model's weights openly and still control who gets access to its more sensitive capabilities?

Tiered Language Models keep those capabilities in the weights but gate access behind a secret key.

Check out this approach to safe open-weight release in our paper:

Excited to share our latest work on Tiered Language Models:

We propose a way to release 100% LLM weights, while keeping specific capabilities locked behind a secret key. It's highly efficient, secure, and a step toward safe open-weight releases.

@gspandana Interesting approach.

It sounds like a potential middle ground between full open-source and controlled access.