Generation on SQuAD
0.837AccuracyCurvZO
Evaluation results
| Method | Links | |
|---|---|---|
| CurvZOModel=OPT-6.7B, Number of training samples=1,000, Optimization Method=CurvZO, Fine-tuning Protocol=Full fine-tuning2026.03 | 0.837 | |
| MeZOModel=OPT-6.7B, Number of training samples=1,000, Optimization Method=MeZO, Fine-tuning Protocol=Full fine-tuning2026.03 | 0.816 | |
| CurvZO + LoRAModel=OPT-6.7B, Number of training samples=1,000, Optimization Method=CurvZO, Fine-tuning Protocol=LoRA2026.03 | 0.806 | |
| MeZO + LoRAModel=OPT-6.7B, Number of training samples=1,000, Optimization Method=MeZO, Fine-tuning Protocol=LoRA2026.03 | 0.801 | |
| DiZO + LoRAModel=OPT-6.7B, Number of training samples=1,000, Optimization Method=DiZO, Fine-tuning Protocol=LoRA2026.03 | 0.715 | |
| DiZOModel=OPT-6.7B, Number of training samples=1,000, Optimization Method=DiZO, Fine-tuning Protocol=Full fine-tuning2026.03 | 0.678 |