Researcher at ArXiv
Author of "Taming the Chaos: Coordinated Autoscaling for Heterogeneous and Disaggregated LLM Inference" in ArXiv
How media typically covers Rongzhi Li
Rongzhi Li as author
ByteDance's HeteroScale autoscaling framework increases GPU utilization by 26.6 percentage points and saves hundreds of thousands of GPU-hours daily in disaggregated LLM inference at scale.
“Author of "Taming the Chaos: Coordinated Autoscaling for Heterogeneous and Disaggregated LLM Inference" in ArXiv”