ZAYA1-8B Matches DeepSeek-R1 on Math with Less Than 1B Active Parameters.
Summary
The article introduces ZAYA1-8B, a sub-1B active-parameter, open-source AI model that reportedly matches larger models on math benchmarks and competes in reasoning and coding tasks. It emphasizes AMD hardware over NVIDIA and describes a Markovian RSA inference approach, highlighting implications for open-source AI and alternative hardware paths for frontier models.