Question Answering on SQuAD (F1, Loss)
0.22LossZO Fine-tuner
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| ZO Fine-tunerModel=Qwen2.5-14B2025.10 | 0.22 | 91 | |
| MeZOModel=Qwen2.5-14B2025.10 | 0.24 | 88 | |
| HIZOOModel=Qwen2.5-14B2025.10 | 0.24 | 89 | |
| LOZOModel=Qwen2.5-14B2025.10 | 0.26 | 90 | |
| MeZO-AdamUModel=Qwen2.5-14B2025.10 | 0.28 | 90 | |
| ZO Fine-tunerModel=LLaMA-3.1-8B2025.10 | 0.31 | 90 | |
| MeZOModel=LLaMA-3.1-8B2025.10 | 0.32 | 89 | |
| HIZOOModel=LLaMA-3.1-8B2025.10 | 0.32 | 88 | |
| LOZOModel=LLaMA-3.1-8B2025.10 | 0.33 | 90 | |
| MeZO-AdamUModel=LLaMA-3.1-8B2025.10 | 0.35 | 86 | |
| ZO Fine-tunerModel=LLaMA-3.2-1B2025.10 | 0.37 | 78 | |
| MeZO-AdamModel=LLaMA-3.2-1B2025.10 | 0.41 | 78 | |
| HIZOOModel=LLaMA-3.2-1B2025.10 | 0.43 | 75 | |
| LOZOModel=LLaMA-3.2-1B2025.10 | 0.47 | 78 | |
| MeZOModel=LLaMA-3.2-1B2025.10 | 0.48 | 75 | |
| ZO Fine-tunerModel=OPT-30B2025.10 | 0.56 | 77 | |
| MeZOModel=OPT-30B2025.10 | 0.59 | 74 | |
| HIZOOModel=OPT-30B2025.10 | 0.65 | 75 | |
| MeZO-AdamUModel=OPT-30B2025.10 | 0.67 | 73 | |
| LOZOModel=OPT-30B2025.10 | 0.96 | 75 |