Vision-Language Driving Reasoning on MAPLM
72.24PerformanceDriveMM
Evaluation Results
| Method | Links | |
|---|---|---|
| DriveMMPrefilling Latency (ms)=1182, Decoding Latency (ms)=61.7, FLOPs=100%, GPU Peak Memory (GB)=16.24, Throughput (samples/s)=1.72, Pruning Setting=None2026.06 | 72.24 | |
| MVPrunerPrefilling Speedup (ms)=299.1(↑3.95×), Decoding Speedup (ms)=51.6(↑1.20×), FLOPs=15.2%, GPU Peak Memory (GB)=15.88, Throughput (samples/s)=0.64(↑2.69×), Pruning Setting=90%2026.06 | 68.3 | |
| Prune2DrivePrefilling Speedup (ms)=274.9(↑4.30×), Decoding Speedup (ms)=55.6(↑1.11×), FLOPs=15.6%, GPU Peak Memory (GB)=15.91, Throughput (samples/s)=0.69(↑2.49×), Pruning Setting=90%2026.06 | 64.81 | |
| SparseVLMPrefilling Speedup (ms)=353.9(↑3.34×), Decoding Speedup (ms)=59.9(↑1.03×), FLOPs=15.8%, GPU Peak Memory (GB)=15.93, Throughput (samples/s)=1.11(↑1.55×), Pruning Setting=90%2026.06 | 56.4 | |
| FastVPrefilling Speedup (ms)=286.9(↑4.12×), Decoding Speedup (ms)=58.2(↑1.06×), FLOPs=15.6%, GPU Peak Memory (GB)=15.91, Throughput (samples/s)=0.82(↑2.10×), Pruning Setting=90%2026.06 | 54.61 |