Kan Zhu

Tactic: Adaptive Sparse Attention with Clustering and Distribution Fitting for Long-Context LLMs

Fiddler: CPU-GPU Orchestration for Fast Inference of Mixture-of-Experts Models