Reasoning on LongReason (val)
79.3Accuracy (val)FA3
Evaluation Results
| Method | Links | |
|---|---|---|
| FA3mb/GPU=4, Peak Mem (GB)=106†, Step (s)=1067 (0.95×), MFU (%)=32.82026.05 | 79.3 | |
| DualKVmb/GPU=4, Peak Mem (GB)=81, Step (s)=687 (1.48×), MFU (%)=58.82026.05 | 78 | |
| DualKVmb/GPU=8, Peak Mem (GB)=93, Step (s)=622 (1.64×), MFU (%)=75.82026.05 | 78 | |
| FA2mb/GPU=4, Peak Mem (GB)=106†, Step (s)=1017 (1.00×), MFU (%)=35.82026.05 | 77.7 |