Commonsense Reasoning on Commonsense Reasoning Benchmark Qwen2.5-1.5B-Instruct (test)
75.82ARC-c AccuracyFSLoRA
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FSLoRABackbone=Qwen2.5-1.5B-Instruct2025.01 | 75.82 | 88.16 | 65.08 | 74.93 | 79.82 | 80.72 | 72.62 | 71.29 | 76.06 | |
| FlexLoRABackbone=Qwen2.5-1.5B-Instruct2025.01 | 73.41 | 88.07 | 63.32 | 71.3 | 76.44 | 78.02 | 70.94 | 68.56 | 73.76 | |
| HeteroLoRABackbone=Qwen2.5-1.5B-Instruct2025.01 | 70.36 | 86.53 | 65.5 | 68.55 | 79.01 | 74.93 | 68.25 | 68.04 | 72.65 | |
| FLoRABackbone=Qwen2.5-1.5B-Instruct2025.01 | 70.28 | 84.51 | 60.9 | 68.82 | 69.58 | 73.45 | 67.53 | 66.27 | 70.17 |