Language Understanding on LLM Benchmark Suite (MMLU, ARC-C, PIQA, WinoG, GSM8K, HellaSwag, GPQA, RACE) (test)
57.93Overall AccuracyBase
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| BaseModel=LLaDA, Sparsity=Dense2026.02 | 57.93 | 65.97 | 43 | 74.1 | 69.3 | 69.29 | 72.7 | 30.4 | 38.7 | |
| Sink-AwareModel=LLaDA, Sparsity=0.25, Base Criterion=SparseGPT2026.02 | 57.68 | 65.33 | 43.09 | 74.37 | 69.53 | 68.58 | 71.98 | 30.12 | 38.4 | |
| Sink-AwareModel=LLaDA, Sparsity=0.25, Base Criterion=Wanda2026.02 | 57.62 | 65.41 | 43.52 | 74.97 | 68.59 | 68.16 | 72.3 | 29.7 | 38.32 | |
| WandaModel=LLaDA, Sparsity=0.252026.02 | 57.43 | 65.2 | 43.94 | 75.3 | 68.59 | 66.03 | 71.95 | 29.85 | 38.55 | |
| SparseGPTModel=LLaDA, Sparsity=0.252026.02 | 57.23 | 65.16 | 43.09 | 74.43 | 67.56 | 67.17 | 72.1 | 30.25 | 38.1 | |
| Sink-AwareModel=LLaDA, Sparsity=0.50, Base Criterion=Wanda2026.02 | 53.18 | 62.16 | 41.38 | 73.18 | 65.27 | 55.88 | 67.18 | 27.95 | 32.45 | |
| WandaModel=LLaDA, Sparsity=0.502026.02 | 52.7 | 61.43 | 39.08 | 72.63 | 64.56 | 57.01 | 67.52 | 27.15 | 32.2 | |
| Sink-AwareModel=LLaDA, Sparsity=0.50, Base Criterion=SparseGPT2026.02 | 52.36 | 60.79 | 39.59 | 72.95 | 65.82 | 52.11 | 67.35 | 27.48 | 32.82 | |
| SparseGPTModel=LLaDA, Sparsity=0.502026.02 | 52.34 | 60.97 | 39.68 | 72.2 | 64.64 | 53.53 | 66.9 | 27.7 | 33.1 | |
| SparseGPTModel=LLaDA, Sparsity=0.752026.02 | 32.57 | 28.6 | 20.99 | 61.75 | 50.04 | 1.52 | 48.2 | 23.9 | 25.55 | |
| Sink-AwareModel=LLaDA, Sparsity=0.75, Base Criterion=SparseGPT2026.02 | 32.57 | 28.93 | 21.08 | 60.12 | 51.07 | 1.9 | 48.7 | 23.55 | 25.2 | |
| Sink-AwareModel=LLaDA, Sparsity=0.75, Base Criterion=Wanda2026.02 | 30.94 | 24.01 | 18.77 | 59.96 | 49.17 | 1.52 | 46.85 | 23.1 | 24.1 | |
| WandaModel=LLaDA, Sparsity=0.752026.02 | 29.99 | 24.76 | 18.52 | 56.69 | 47.43 | 0.99 | 45.25 | 22.85 | 23.45 |