ZAYA1-8B-Diffusion-Preview
ZAYA1-8B-Diffusion-Preview is the first mixture-of-experts diffusion language model converted from an autoregressive LLM and the first diffusion LM trained on AMD GPUs. It generates 16 tokens per step, achieving up to 7.7× inference speedup over the autoregressive base, using Zyphra's TiDAR recipe and CCA attention.