Long-context Understanding on RULER 8K
70.98AccuracyFull Attention
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Full AttentionModel=Dream2026.05 | 70.98 | 73.56 | |
| PulseColModel=Dream, Attention Sparsity=50%2026.05 | 70.77 | 74.08 | |
| PulseColModel=Dream, Attention Sparsity=80%2026.05 | 70.53 | 72.83 | |
| SparseDModel=Dream, Attention Sparsity=50%2026.05 | 66.16 | 70.85 | |
| SparseDModel=Dream, Attention Sparsity=80%2026.05 | 61.76 | 49.85 | |
| SparseDModel=LLaDA-1.5, Attention Sparsity=80%2026.05 | 61.38 | 63.89 | |
| SparseDModel=LLaDA-1.5, Attention Sparsity=50%2026.05 | 61.34 | 67.27 | |
| Full AttentionModel=LLaDA-1.52026.05 | 60.22 | 67.58 | |
| PulseColModel=LLaDA-1.5, Attention Sparsity=50%2026.05 | 60.06 | 67.47 | |
| PulseColModel=LLaDA-1.5, Attention Sparsity=80%2026.05 | 59.98 | 66.58 | |
| StreamingLLMModel=LLaDA-1.52026.05 | 38.38 | 44.9 | |
| StreamingLLMModel=Dream2026.05 | 36.91 | 53.23 | |
| Sliding WindowModel=LLaDA-1.52026.05 | 32.65 | 39.58 | |
| Sliding WindowModel=Dream2026.05 | 31.36 | 48.49 |