SOLARIS:Speculative Offloading of Latent-based Representation for Inference Scaling
Zikun Liu · Liang Luo · Qianru Li · Zhengyu Zhang · Wei Ling · Jingyi Shen · Zeliang Chen · Yaning Huang · Jingxian Huang · Abdallah Aboelela · Chonglin Sun · Feifan Gu · Fenggang Wu · Hang Qu · Huayu Li · Jill Pan · Kaidi Pei · Laming Chen · Longhao Jin · Qin Huang · Tongyi Tang · Varna Puvvada · Wenlin Chen · Xiaohan Wei · Xu Cao · Yantao Yao · Yuan Jin · Yunchen Pu · Yuxin Chen · Zijian Shen · Zhengkai Zhang · Dong Liang · Ellie Dingqiao Wen
摘要
Recent advances in recommendation scaling laws have led to foundation models of unprecedented complexity. While these models offer superior performance, their computational demands make real-time serving impractical, often forcing practitioners to rely on knowledge distillation—compromising servi...