Mathematical Reasoning on MATH (test) (Accuracy + Throughput + Speedup)
39.8AccuracyDream
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| DreamGen Length=512, Evaluation Protocol=4-shot2026.01 | 39.8 | 8.6 | 1 | |
| Streaming-dLLMGen Length=512, Evaluation Protocol=4-shot2026.01 | 39.4 | 96 | 11.2 | |
| Fast-dLLMGen Length=512, Evaluation Protocol=4-shot2026.01 | 39.3 | 54.4 | 6.3 | |
| dKV-CacheGen Length=512, Evaluation Protocol=4-shot2026.01 | 38.5 | 11.6 | 1.3 | |
| DreamGen Length=256, Evaluation Protocol=4-shot2026.01 | 38.4 | 10.5 | 1 | |
| Prefix-CacheGen Length=512, Evaluation Protocol=4-shot2026.01 | 38 | 24.5 | 2.8 | |
| Fast-dLLMGen Length=256, Evaluation Protocol=4-shot2026.01 | 37.6 | 62.6 | 5.98 | |
| Streaming-dLLMGen Length=256, Evaluation Protocol=4-shot2026.01 | 37.6 | 78.4 | 7.5 | |
| dKV-CacheGen Length=256, Evaluation Protocol=4-shot2026.01 | 36.8 | 14.6 | 1.4 | |
| Prefix-CacheGen Length=256, Evaluation Protocol=4-shot2026.01 | 36.8 | 32.5 | 3.1 |