5 papers
DraftExpert: Expansion-Aware Self-Speculative Decoding for End-Device MoE Inference
Dengke Han
Large Mixture-of-Experts (MoE) language models are attractive for end-device deployment because only a small subset of experts is active per token, but their routed expert weights…
A Systematic Characterization of LLM Inference on GPUs
Haonan Wang, Xuxin Xiao, Mingyu Yan +8
This work presents a systematic characterization of Large Language Model (LLM) inference to address fragmented understanding. Through comprehensive experiments, we establish a four…
TLV-HGNN: Thinking Like a Vertex for Memory-efficient HGNN Inference
Dengke Han, Duo Wang, Mingyu Yan +2
Heterogeneous graph neural networks (HGNNs) excel at processing heterogeneous graph data and are widely applied in critical domains. In HGNN inference, the neighbor aggregation sta…
Accelerating GNN Training through Locality-aware Dropout and Merge
Gongjian Sun, Mingyu Yan, Dengke Han +4
Graph Neural Networks (GNNs) have demonstrated significant success in graph learning and are widely adopted across various critical domains. However, the irregular connectivity bet…
Characterizing and Understanding HGNN Training on GPUs
Dengke Han, Mingyu Yan, Xiaochun Ye +1
Owing to their remarkable representation capabilities for heterogeneous graph data, Heterogeneous Graph Neural Networks (HGNNs) have been widely adopted in many critical real-world…