Latest

• FlowPrefill: Decoupling Preemption from Prefill Scheduling Granularity to Mitigate Head-of-Line Blocking in LLM Serving
• A Flexible Programmable Pipeline Parallelism Framework for Efficient DNN Training
• TraceFlow: Efficient Trace Analysis for Large-Scale Parallel Applications via Interaction Pattern-Aware Trace Distribution
• Training Large Models on Heterogeneous and Geo-Distributed Resource with Constricted Networks
• UltraAttn: Efficiently Parallelizing Attention through Hierarchical Context-Tiling
• SmartMoE: Efficiently Training Sparsely-Activated Models through Combining Offline and Online Parallelization