kyegomez/OpenMythos
An open-source, theoretical reconstruction of the Claude Mythos LLM architecture, featuring a Recurrent-Depth Transformer and sparse Mixture of Experts for advanced reasoning.
Core Features
Quick Start
pip install open-mythosDetailed Introduction
OpenMythos is an independent, community-driven open-source project that aims to theoretically reconstruct the Claude Mythos large language model architecture based on publicly available research. It introduces a novel Recurrent-Depth Transformer (RDT) design, incorporating distinct Prelude, Recurrent Block, and Coda stages. A key innovation is its sparse Mixture of Experts (MoE) implementation, enabling efficient, compute-adaptive, and depth-variable reasoning. Developed in PyTorch, OpenMythos provides researchers and developers with a flexible framework to explore and experiment with advanced LLM architectures, offering scalable model configurations from 1 billion to 1 trillion parameters.