DF-GNN speeds up attention graph neural network training on GPUs with dynamic thread scheduling and two fusion strategies, reporting up to 7x kernel speedup over DGL sparse.
Do transformers really perform badly for graph representation? Advances in Neural Information Processing Systems, 34:28877–28888, 2021
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
DF-GNN: Dynamic Fusion Framework for Attention Graph Neural Networks on GPUs
DF-GNN speeds up attention graph neural network training on GPUs with dynamic thread scheduling and two fusion strategies, reporting up to 7x kernel speedup over DGL sparse.