XQuant: Achieving Ultra-Low Bit KV Cache Quantization with Cross-Layer CompressionHaoqi Yang, Yao Yao, Zuchao Li, Baoyuan Qi, Liu Guoming, Hai Zhao
Rethinking Fixed-Budget Expert Activation in Fine-Grained MoEsHaoqi Yang*, Luohe Shi*, Qiwei Li, Zuchao Li, Ping Wang, Hao Huang, Hai Zhao
ROCOB: Dual-Manifold Alignment with Optimal Budgeting for Training-Free PTQ RecoveryHaoqi Yang, Teng Xiao, Zuchao Li, Hao Huang, Lefei Zhang, Hai Zhao