Liquid AI opens previews of its first Liquid Foundation Models

Liquid AI introduced 1.3B, 3.1B and 40.3B language models with hosted playground access. The initial Liquid Foundation Models explored alternatives to transformer architectures; weights were not publicly released.

Hosted access to the first generation

The initial models could be tested through hosted services. Liquid AI described an architecture drawing on dynamical systems and signal processing. Its efficiency comparisons were vendor results, and this announcement did not release model weights. Later LFM2 checkpoints had a different hybrid architecture and separate licensing. [1]

Sources

  1. Liquid Foundation Models: Our First Series of Generative AI Models

    Liquid AI · Sep 30, 2024

  2. MIT spinoff Liquid debuts non-transformer AI models and they're already state-of-the-art

    VentureBeat · Sep 30, 2024