Provides an open-source, theoretical implementation of the Claude Mythos model architecture based on available research.
This project theoretically reconstructs the Claude Mythos architecture using a Recurrent-Depth Transformer (RDT) with Prelude, a looped Recurrent Block, and Coda stages. It includes switchable attention (MLA, GQA) and a sparse Mixture-of-Experts for compute-adaptive reasoning. Install it with `pip install open-mythos` and use by importing components from `open_mythos.main`.
Provides an open-source, theoretical implementation of the Claude Mythos model architecture based on available research.
AI researchers and developers keen on exploring or extending advanced theoretical transformer architectures, especially those inspired by Anthropic's Claude, will find this valuable.