Arithmetic Reasoning on GSM8K and MATH Average
43.4Average AccuracyDiaBlo
Evaluation Results
| Method | Links | |
|---|---|---|
| DiaBlor/N=N = 32, #Params=2.09%, Base Model=Llama2-7B, Evaluation Protocol=Fine-tuning2025.06 | 43.4 | |
| Full FTr/N=N/A, #Params=100%, Base Model=Llama2-7B, Evaluation Protocol=Fine-tuning2025.06 | 43.2 | |
| DiaBlor/N=N = 64, #Params=1.04%, Base Model=Llama2-7B, Evaluation Protocol=Fine-tuning2025.06 | 42.1 | |
| MiLoRAr/N=r = 64, #Params=1.67%, Base Model=Llama2-7B, Evaluation Protocol=Fine-tuning2025.06 | 40.7 | |
| LoRAr/N=r = 64, #Params=1.67%, Base Model=Llama2-7B, Evaluation Protocol=Fine-tuning2025.06 | 38.7 | |
| PiSSAr/N=r = 64, #Params=1.67%, Base Model=Llama2-7B, Evaluation Protocol=Fine-tuning2025.06 | 37 | |
| Zero-shotr/N=N/A, #Params=0%, Base Model=Llama2-7B, Evaluation Protocol=Zero-shot2025.06 | 1.1 |