Commonsense Reasoning on Commonsense Reasoning Suite (ARC-e, OBQA, SIQA, ARC-c, WinoG., PIQA)
88ARC-e AccuracyLoRA
Evaluation results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| LoRAModel=LLaMA-3.2 3B, Inference Noise=02025.02 | 88 | 82.2 | 76.8 | 76.4 | 77.2 | 83.8 | 80.7 | — | |
| LoRAModel=LLaMA-3.2 3B, Inference Noise=5e-32025.02 | 87.5 | 81.6 | 76.7 | 75.7 | 76.2 | 80.5 | 79.7 | — | |
| HaLoRAModel=LLaMA-3.2 3B, Inference Noise=02025.02 | 87.3 | 81.4 | 77.7 | 76.7 | 80.6 | 84.2 | 81.3 | 0.6 | |
| HaLoRAModel=LLaMA-3.2 3B, Inference Noise=5e-32025.02 | 87 | 82.3 | 77.6 | 76.4 | 80.1 | 83.9 | 81.2 | 1.5 | |
| HaLoRAModel=LLaMA-3.2 3B, Inference Noise=1e-22025.02 | 86.6 | 81.6 | 77.2 | 76 | 79.3 | 83.4 | 80.7 | 2.8 | |
| LoRAModel=LLaMA-3.2 3B, Inference Noise=1e-22025.02 | 86.1 | 80.6 | 76.3 | 73.9 | 71.5 | 79 | 77.9 | — | |
| HaLoRAModel=LLaMA-3.2 3B, Inference Noise=2e-22025.02 | 84.6 | 78.8 | 75.8 | 72.8 | 77.2 | 81 | 78.4 | 13.5 | |
| HaLoRAModel=LLaMA-3.2 1B, Inference Noise=02025.02 | 73.3 | 68 | 69 | 55.5 | 65.4 | 74.6 | 67.6 | 5.3 | |
| HaLoRAModel=LLaMA-3.2 1B, Inference Noise=5e-32025.02 | 73.1 | 66.8 | 69.4 | 54.8 | 64.6 | 74.3 | 67.2 | 6.3 | |
| LoRAModel=LLaMA-3.2 3B, Inference Noise=2e-22025.02 | 72.3 | 65.2 | 69 | 60.1 | 57.5 | 65.5 | 64.9 | — | |
| HaLoRAModel=LLaMA-3.2 1B, Inference Noise=1e-22025.02 | 71.8 | 65.2 | 68.9 | 54.2 | 64.1 | 73.5 | 66.3 | 9.3 | |
| LoRAModel=Qwen2.5 0.5B, Inference Noise=02025.02 | 70.5 | 57 | 64.8 | 53.2 | 56 | 51.4 | 58.8 | — | |
| HaLoRAModel=Qwen2.5 0.5B, Inference Noise=02025.02 | 69.5 | 58.2 | 65 | 52.9 | 56.6 | 68.7 | 61.8 | 3 | |
| LoRAModel=LLaMA-3.2 1B, Inference Noise=02025.02 | 69.5 | 64.4 | 61 | 46.7 | 57.8 | 74.2 | 62.3 | — | |
| LoRAModel=Qwen2.5 0.5B, Inference Noise=5e-32025.02 | 69.1 | 56.4 | 59.2 | 51.8 | 54.9 | 48.7 | 56.7 | — | |
| HaLoRAModel=Qwen2.5 0.5B, Inference Noise=5e-32025.02 | 68.8 | 58.3 | 64.6 | 52.3 | 57.4 | 67.7 | 61.5 | 4.8 | |
| HaLoRAModel=LLaMA-3.2 1B, Inference Noise=2e-22025.02 | 68.6 | 60.2 | 66 | 50.8 | 61.9 | 71.1 | 63.1 | 22.7 | |
| LoRAModel=LLaMA-3.2 1B, Inference Noise=5e-32025.02 | 67.1 | 59.4 | 59.6 | 48.5 | 58.8 | 72 | 60.9 | — | |
| HaLoRAModel=Qwen2.5 0.5B, Inference Noise=1e-22025.02 | 66.1 | 56.3 | 62.1 | 49.8 | 57.2 | 61.3 | 58.8 | 13.9 | |
| LoRAModel=LLaMA-3.2 1B, Inference Noise=1e-22025.02 | 62.2 | 53.2 | 58 | 44.7 | 57.5 | 66.5 | 57 | — | |
| HaLoRAModel=Qwen2.5 0.5B, Inference Noise=2e-22025.02 | 52.6 | 42.7 | 51.2 | 40.3 | 49.1 | 55.6 | 48.6 | 20.5 | |
| LoRAModel=Qwen2.5 0.5B, Inference Noise=1e-22025.02 | 47.5 | 40.7 | 48.7 | 35.9 | 51.6 | 44.7 | 44.9 | — | |
| LoRAModel=LLaMA-3.2 1B, Inference Noise=2e-22025.02 | 40.9 | 38.1 | 46.8 | 31 | 45.1 | 40.7 | 40.4 | — | |
| LoRAModel=Qwen2.5 0.5B, Inference Noise=2e-22025.02 | 23.7 | 18.4 | 23.5 | 18.7 | 46.9 | 37.2 | 28.1 | — |