Ask or search anything...

Ring-lite, developed by Inclusion AI's Ling Team, presents an open-source multi-domain reasoning model based on a Mixture-of-Experts (MoE) architecture, achieving strong performance while activating fewer parameters. It introduces C3PO, an algorithm-system co-design, to stabilize reinforcement learning (RL) training and mitigate 'reward collapse' in LLMs.