Graph Transformer Efficiency Benchmarking on city-roads L
9.7Latency Speedup (Forward)CSR fused kernels
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| CSR fused kernelsHeads (H)=2, Head Dimension (D)=642026.05 | 9.7 | 1.71 | 1.77 | 1.51 | |
| CSR fused kernelsHeads (H)=8, Head Dimension (D)=1282026.05 | 2.01 | 1.34 | 2.37 | 1.53 | |
| PyGHeads (H)=2, Head Dimension (D)=642026.05 | 1.54 | 0.62 | 0.54 | 0.72 | |
| Ours (CSR fused)Number of Heads (H)=4, Head Dimension (D)=1282026.05 | 1.15 | 1.16 | 1.16 | 1 | |
| Ours (WSB/TC)Number of Heads (H)=4, Head Dimension (D)=1282026.05 | 1.15 | 1.03 | 1.07 | 1.21 | |
| DF-GNNNumber of Heads (H)=4, Head Dimension (D)=1282026.05 | 1.13 | 1.18 | 1.16 | 1.13 | |
| PyGHeads (H)=8, Head Dimension (D)=1282026.05 | 0.42 | 0.58 | 0.48 | 0.61 |