OpenMOSS

DiRL: An Efficient Post-Training Framework for Diffusion Language Models

Beyond Real: Imaginary Extension of Rotary Position Embeddings for Long-Context LLMs