Commonsense Reasoning on Commonsense170K
93.9ARC-c AccuracyDoRA
Evaluation results
| Method | Links | |||||||||
|---|---|---|---|---|---|---|---|---|---|---|
| DoRAModel=Phi-4-14B, # Params (%)=0.772025.09 | 93.9 | 98.2 | 87.3 | 82 | 94.8 | 75.1 | 92.4 | 89.9 | 89.2 | |
| LoRAModel=Phi-4-14B, # Params (%)=0.752025.09 | 93.5 | 98 | 87.5 | 81.9 | 93.8 | 74.6 | 92.6 | 95.1 | 89.6 | |
| HyperAdaptModel=Phi-4-14B, # Params (%)=0.022025.09 | 93.3 | 97.7 | 83.1 | 81 | 91.2 | 69.7 | 92.6 | 94.5 | 87.9 | |
| LoRAr=1Model=Phi-4-14B, # Params (%)=0.022025.09 | 92.8 | 97.9 | 83.5 | 79.6 | 91.6 | 73.4 | 90.5 | 94 | 87.9 | |
| VeRAModel=Phi-4-14B, # Params (%)=0.02 | 0.752025.09 | 92.6 | 97.8 | 58.4 | 79.3 | 90.8 | 69.9 | 83.6 | 93.6 | 83.2 | |
| LoRAModel=Qwen-2.5-7B, # Params (%)=1.052025.09 | 88.6 | 95.8 | 83.5 | 80.2 | 89.2 | 72.7 | 89.8 | 94.9 | 86.8 | |
| DoRAModel=Qwen-2.5-7B, # Params (%)=1.072025.09 | 88.5 | 95.9 | 82.4 | 79.8 | 89.6 | 72.8 | 89.6 | 94.6 | 86.7 | |
| HyperAdaptModel=Qwen-2.5-7B, # Params (%)=0.032025.09 | 88.3 | 95.3 | 80.1 | 78.8 | 90.4 | 68.6 | 88.7 | 93.7 | 85.5 | |
| LoRAr=1Model=Qwen-2.5-7B, # Params (%)=0.032025.09 | 88.1 | 95.8 | 76 | 78.9 | 87.4 | 70.1 | 88 | 92.8 | 84.6 | |
| VeRAModel=Qwen-2.5-7B, # Params (%)=0.02 | 0.512025.09 | 88 | 95.4 | 74.8 | 78.6 | 87.8 | 69.2 | 88.4 | 92.6 | 84.3 | |
| DoRAModel=Llama-3-8B, # Params (%)=1.052025.09 | 79.6 | 90.8 | 83.8 | 80.1 | 84.2 | 73.2 | 87.9 | 95.5 | 84.4 | |
| LoRAModel=Llama-3-8B, # Params (%)=1.032025.09 | 79.4 | 90.3 | 83 | 79.8 | 86 | 72.5 | 87.9 | 95.5 | 84.3 | |
| HyperAdaptModel=Llama-3-8B, # Params (%)=0.032025.09 | 78.2 | 89.3 | 79.4 | 76.4 | 80.6 | 67.9 | 86.3 | 92.4 | 81.3 | |
| LoRAr=1Model=Llama-3-8B, # Params (%)=0.032025.09 | 76.5 | 89.7 | 77.4 | 75.2 | 78.8 | 60.8 | 84.9 | 89.9 | 79.1 | |
| VeRAModel=Llama-3-8B, # Params (%)=0.02 | 0.372025.09 | 74.4 | 89 | 74 | 73.3 | 78.8 | 61.6 | 84 | 84.5 | 77.5 |