Commonsense Reasoning on Commonsense Reasoning LLaMA-3.2-3B-Instruct (test)
76.1ARC-cFSLoRA
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| FSLoRABackbone=LLaMA-3.2-3B-Instruct, GPU hours=44.3h2025.01 | 76.1 | 87.2 | 69.3 | 82.2 | 80.7 | 84 | 76.8 | 79.1 | 79.4 | |
| FlexLoRABackbone=LLaMA-3.2-3B-Instruct, GPU hours=68.3h2025.01 | 74.2 | 86.7 | 68.6 | 79.4 | 75.8 | 81 | 75.9 | 77.9 | 77.4 | |
| HeteroLoRABackbone=LLaMA-3.2-3B-Instruct, GPU hours=43.7h2025.01 | 73.4 | 86.6 | 65.8 | 73 | 71.4 | 80.9 | 73.8 | 72 | 74.6 | |
| RAVANBackbone=LLaMA-3.2-3B-Instruct, GPU hours=51.2h2025.01 | 71.9 | 81.6 | 65.7 | 76.7 | 75.6 | 77.1 | 74.8 | 73.8 | 74.7 | |
| FLoRABackbone=LLaMA-3.2-3B-Instruct, GPU hours=49.8h2025.01 | 68.3 | 83.1 | 65.8 | 77.2 | 74.2 | 80.5 | 76.1 | 71.5 | 74.6 | |
| Base modelBackbone=LLaMA-3.2-3B-Instruct2025.01 | 51.5 | 61.4 | 47.2 | 55.8 | 52.4 | 21.3 | 54.4 | 4.6 | 43.6 |