Question Answering on BoolQ (Loss and Accuracy)
0.23LossMeZO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| MeZOModel=Qwen2.5-14B2025.10 | 0.23 | 84 | |
| HIZOOModel=Qwen2.5-14B2025.10 | 0.25 | 86 | |
| ZO Fine-tunerModel=Qwen2.5-14B2025.10 | 0.29 | 89 | |
| LOZOModel=Qwen2.5-14B2025.10 | 0.32 | 86 | |
| ZO Fine-tunerModel=LLaMA-3.1-8B2025.10 | 0.34 | 87 | |
| MeZO-AdamUModel=Qwen2.5-14B2025.10 | 0.35 | 84 | |
| HIZOOModel=LLaMA-3.1-8B2025.10 | 0.36 | 79 | |
| LOZOModel=LLaMA-3.1-8B2025.10 | 0.41 | 83 | |
| MeZOModel=LLaMA-3.1-8B2025.10 | 0.42 | 78 | |
| MeZO-AdamUModel=LLaMA-3.1-8B2025.10 | 0.5 | 75 | |
| ZO Fine-tunerModel=LLaMA-3.2-1B2025.10 | 0.58 | 66 | |
| MeZOModel=OPT-30B2025.10 | 0.6 | 66 | |
| ZO Fine-tunerModel=OPT-30B2025.10 | 0.61 | 67 | |
| HIZOOModel=LLaMA-3.2-1B2025.10 | 0.62 | 61 | |
| LOZOModel=LLaMA-3.2-1B2025.10 | 0.62 | 64 | |
| MeZO-AdamUModel=OPT-30B2025.10 | 0.62 | 66 | |
| HIZOOModel=OPT-30B2025.10 | 0.62 | 65 | |
| MeZOModel=LLaMA-3.2-1B2025.10 | 0.63 | 63 | |
| MeZO-AdamModel=LLaMA-3.2-1B2025.10 | 0.66 | 62 | |
| LOZOModel=OPT-30B2025.10 | 0.7 | 66 |