Throughput, Latency, and Score on GPQA (Science Question Answering)
34.6ScoredLLM-Cache
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| dLLM-CacheBackbone=Dream-Base, Few-shot=0-shot, Generation Length=2562025.09 | 34.6 | 11.12 | 23.1 | |
| Dream-Instruct 7BSource=Originally Published2025.10 | 33 | — | — | |
| VanillaBackbone=Dream-Base, Few-shot=0-shot, Generation Length=2562025.09 | 32.8 | 6.54 | 40.55 | |
| Dream-Instruct 7B + G-StarRefinement Strategy=Loop-based refinement2025.10 | 32.8 | — | — | |
| d2CacheBackbone=Dream-Base, Few-shot=0-shot, Generation Length=2562025.09 | 32.6 | 13.02 | 18.64 | |
| Fast-dLLMBackbone=Dream-Base, Few-shot=0-shot, Generation Length=2562025.09 | 31.5 | 11.92 | 21.45 | |
| dLLM-CacheBackbone=LLaDA-Base, Few-shot=0-shot, Generation Length=2562025.09 | 31 | 11.32 | 22.69 | |
| Fast-dLLMBackbone=LLaDA-Base, Few-shot=0-shot, Generation Length=2562025.09 | 31 | 12.67 | 20.24 | |
| Dream-Instruct 7BSource=Reproduced2025.10 | 31 | — | — | |
| d2CacheBackbone=LLaDA-Base, Few-shot=0-shot, Generation Length=2562025.09 | 30.8 | 15.32 | 16.77 | |
| VanillaBackbone=LLaDA-Base, Few-shot=0-shot, Generation Length=2562025.09 | 30.4 | 6.27 | 42.35 |