Reasoning on MMLU GSM8K
72.93MMLU AccuracyFP16
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| FP16Model=Qwen3 8B2026.05 | 72.93 | 86.05 | 79.49 | |
| SOARModel=Qwen3 8B2026.05 | 71.47 | 82.56 | 77.02 | |
| RaZeRModel=Qwen3 8B2026.05 | 71.31 | 82.41 | 76.86 | |
| NVFP4Model=Qwen3 8B2026.05 | 70.7 | 79.3 | 75 | |
| 4over6Model=Qwen3 8B2026.05 | 70.61 | 81.96 | 76.29 | |
| FP16Model=LLaMA-3.2 3B-Instruct2026.05 | 62.3 | 77.1 | 69.7 | |
| RaZeRModel=LLaMA-3.2 3B-Instruct2026.05 | 59.29 | 72.93 | 66.11 | |
| SOARModel=LLaMA-3.2 3B-Instruct2026.05 | 59.23 | 73.62 | 66.43 | |
| NVFP4Model=LLaMA-3.2 3B-Instruct2026.05 | 59.07 | 71.19 | 65.13 | |
| 4over6Model=LLaMA-3.2 3B-Instruct2026.05 | 58.43 | 71.8 | 65.12 | |
| FP16Model=LLaMA-3.2 1B-Instruct2026.05 | 48.25 | 45.26 | 46.76 | |
| SOARModel=LLaMA-3.2 1B-Instruct2026.05 | 44.05 | 32.83 | 38.44 | |
| 4over6Model=LLaMA-3.2 1B-Instruct2026.05 | 43.98 | 30.93 | 37.46 | |
| RaZeRModel=LLaMA-3.2 1B-Instruct2026.05 | 43.74 | 32.22 | 37.98 | |
| NVFP4Model=LLaMA-3.2 1B-Instruct2026.05 | 43.28 | 31.77 | 37.53 |