Inference Throughput on AIME24/25
5,888Throughput (token/s)Mix-RL-4B
Evaluation results
| Method | Links | |
|---|---|---|
| Mix-RL-4BMode=Fold, Memory Limit=48Gb2026.02 | 5,888 | |
| Fold-RL-4BMode=Fold, Memory Limit=48Gb2026.02 | 5,612 | |
| Mix-RL-4BMode=Fold, Memory Limit=24Gb2026.02 | 3,182 | |
| Fold-RL-4BMode=Fold, Memory Limit=24Gb2026.02 | 2,971 | |
| Unfold-RL-4BMode=Unfold, Memory Limit=48Gb2026.02 | 1,483 | |
| Unfold-RL-4BMode=Unfold, Memory Limit=24Gb2026.02 | 1,083 |