Natural Language Inference on CB (Loss, Acc)
0.03LossHIZOO
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| HIZOOModel=Qwen2.5-14B2025.10 | 0.03 | 89 | |
| ZO Fine-tunerModel=Qwen2.5-14B2025.10 | 0.03 | 93 | |
| MeZOModel=Qwen2.5-14B2025.10 | 0.05 | 93 | |
| MeZO-AdamUModel=Qwen2.5-14B2025.10 | 0.13 | 91 | |
| LOZOModel=Qwen2.5-14B2025.10 | 0.19 | 91 | |
| ZO Fine-tunerModel=LLaMA-3.1-8B2025.10 | 0.26 | 76 | |
| HIZOOModel=LLaMA-3.2-1B2025.10 | 0.44 | 71 | |
| HIZOOModel=LLaMA-3.1-8B2025.10 | 0.47 | 71 | |
| MeZO-AdamModel=LLaMA-3.2-1B2025.10 | 0.53 | 66 | |
| MeZOModel=LLaMA-3.1-8B2025.10 | 0.54 | 71 | |
| LOZOModel=LLaMA-3.1-8B2025.10 | 0.54 | 70 | |
| MeZOModel=LLaMA-3.2-1B2025.10 | 0.55 | 70 | |
| ZO Fine-tunerModel=LLaMA-3.2-1B2025.10 | 0.57 | 67 | |
| LOZOModel=LLaMA-3.2-1B2025.10 | 0.58 | 64 | |
| MeZO-AdamUModel=LLaMA-3.1-8B2025.10 | 0.61 | 70 | |
| ZO Fine-tunerModel=OPT-30B2025.10 | 0.66 | 70 | |
| HIZOOModel=OPT-30B2025.10 | 0.67 | 66 | |
| MeZOModel=OPT-30B2025.10 | 0.69 | 64 | |
| MeZO-AdamUModel=OPT-30B2025.10 | 0.7 | 66 | |
| LOZOModel=OPT-30B2025.10 | 0.73 | 64 |