Mathematical Reasoning on MATH (test) (Accuracy + Latency + Speedup)
2.7Latency (s)Streaming-dLLM
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Streaming-dLLMGen Length=256, Evaluation Protocol=4-shot2026.01 | 2.7 | 33.7 | 10.3 | — | |
| Streaming-dLLMGen Length=256, Prompting Strategy=4-shot2026.01 | 3.2 | 37.6 | 7.7 | — | |
| Fast-dLLMGen Length=256, Prompting Strategy=4-shot2026.01 | 4.2 | 37.6 | 5.8 | — | |
| Streaming-dLLMGen Length=512, Evaluation Protocol=4-shot2026.01 | 4.6 | 35.1 | 14.4 | — | |
| Fast-dLLMGen Length=256, Evaluation Protocol=4-shot2026.01 | 4.7 | 32.6 | 5.9 | — | |
| Streaming-dLLMGen Length=512, Prompting Strategy=4-shot2026.01 | 5.2 | 39.4 | 11.4 | — | |
| Prefix-CacheGen Length=256, Prompting Strategy=4-shot2026.01 | 7.9 | 36.8 | 3.1 | — | |
| Fast-dLLMGen Length=512, Evaluation Protocol=4-shot2026.01 | 8.4 | 35.1 | 7.9 | — | |
| Prefix-CacheGen Length=256, Evaluation Protocol=4-shot2026.01 | 8.5 | 32.5 | 3.3 | — | |
| Fast-dLLMGen Length=512, Prompting Strategy=4-shot2026.01 | 9.4 | 39.3 | 6.3 | — | |
| dKV-CacheGen Length=256, Prompting Strategy=4-shot2026.01 | 17.5 | 36.8 | 1.4 | — | |
| dKV-CacheGen Length=256, Evaluation Protocol=4-shot2026.01 | 17.6 | 31.8 | 1.6 | — | |
| Prefix-CacheGen Length=512, Prompting Strategy=4-shot2026.01 | 20.9 | 38 | 2.8 | — | |
| Prefix-CacheGen Length=512, Evaluation Protocol=4-shot2026.01 | 22.9 | 35 | 2.9 | — | |
| DreamGen Length=256, Prompting Strategy=4-shot2026.01 | 24.5 | 38.4 | 1 | — | |
| LLaDA-1.5Gen Length=256, Evaluation Protocol=4-shot2026.01 | 27.9 | 32.7 | 1 | — | |
| dKV-CacheGen Length=512, Evaluation Protocol=4-shot2026.01 | 42.1 | 35.1 | 1.6 | — | |
| dKV-CacheGen Length=512, Prompting Strategy=4-shot2026.01 | 44.1 | 38.5 | 1.3 | — | |
| DreamGen Length=512, Prompting Strategy=4-shot2026.01 | 59.3 | 39.8 | 1 | — | |
| LLaDA-1.5Gen Length=512, Evaluation Protocol=4-shot2026.01 | 66.3 | 37.1 | 1 | — | |
| MSVN=42026.03 | 69.2 | — | — | 21.8 | |
| MSVN=162026.03 | 71.6 | — | — | 59.6 | |
| ProbeN=42026.03 | 88.5 | — | — | — | |
| ProbeN=162026.03 | 177.3 | — | — | — | |
| MSVN=642026.03 | 199 | — | — | 65.4 | |
| ProbeN=642026.03 | 575 | — | — | — |