Classification on WSC
80AccuracyFull Fine-Tuning (FT)
Evaluation results
| Method | Links | |
|---|---|---|
| Full Fine-Tuning (FT)Model=Qwen2.5-14B, Trainable Parameters=14.7B2025.12 | 80 | |
| LoRAModel=Qwen2.5-14B, Trainable Parameters=6.3M2025.12 | 76 | |
| Partial-LoRAModel=Qwen2.5-14B, Trainable Parameters=400k2025.12 | 74.1 | |
| Hybrid-LoRAModel=Vicuna-7b-v1.52026.04 | 72.1 | |
| First Order Adamw FTModel Backbone=Llama2-7B, Fine-tuning Strategy=Full Fine-tuning, Optimization Algorithm=First Order Adamw, Number of training examples=10002026.06 | 71.15 | |
| Masking (0.001%)Model=Qwen2.5-14B, Trainable Parameters=149K2025.12 | 70.2 | |
| Partial-LoRAModel=LLAMA2-7B, Trainable Parameters=50k2025.12 | 66.4 | |
| LoRAModel=LLAMA3.1-8B, Trainable Parameters=3.4M2025.12 | 66.4 | |
| Partial-LoRAModel=LLAMA3.1-8B, Trainable Parameters=200k2025.12 | 66.4 | |
| MeZO-GV (LoRA)Optimization Algorithm=MeZO, Tuning Strategy=LoRA, Guiding Vectors=true, Model=OPT-13B2026.01 | 66.3 | |
| Hybrid-PrefixModel=Vicuna-7b-v1.52026.04 | 66.3 | |
| SubZero (FT)Optimization Algorithm=SubZero, Tuning Strategy=Full Tuning, Guiding Vectors=false, Model=OPT-13B2026.01 | 65.4 | |
| MeZO-GV (FT)Optimization Algorithm=MeZO, Tuning Strategy=Full Tuning, Guiding Vectors=true, Model=OPT-13B2026.01 | 65.4 | |
| SubZero (LoRA)Optimization Algorithm=SubZero, Tuning Strategy=LoRA, Guiding Vectors=false, Model=OPT-13B2026.01 | 65.4 | |
| SubZero-GV (LoRA)Optimization Algorithm=SubZero, Tuning Strategy=LoRA, Guiding Vectors=true, Model=OPT-13B2026.01 | 65.4 | |
| Full Fine-Tuning (FT)Model=LLAMA3.1-8B, Trainable Parameters=8B2025.12 | 64.9 | |
| Masking (0.001%)Model=LLAMA3.1-8B, Trainable Parameters=80K2025.12 | 64.5 | |
| Dominant-layer ZO FTModel Backbone=Llama2-7B, Fine-tuning Strategy=Full Fine-tuning, Optimization Algorithm=Dominant-layer ZO, Number of training examples=10002026.06 | 64.42 | |
| SubZero-GV (FT)Optimization Algorithm=SubZero, Tuning Strategy=Full Tuning, Guiding Vectors=true, Model=OPT-13B2026.01 | 64.4 | |
| MeZO (LoRA)Optimization Algorithm=MeZO, Tuning Strategy=LoRA, Guiding Vectors=false, Model=OPT-13B2026.01 | 64.4 | |
| SubZero-GV (Prefix)Optimization Algorithm=SubZero, Tuning Strategy=Prefix Tuning, Guiding Vectors=true, Model=OPT-13B2026.01 | 64.4 | |
| Masking (0.001%)Model=LLAMA2-7B, Trainable Parameters=68K2025.12 | 64.4 | |
| FO-PromptModel=Vicuna-7b-v1.52026.04 | 64.4 | |
| HiZOOOptimization Algorithm=HiZOO, Tuning Strategy=Full Tuning, Model=OPT-13B2026.01 | 63.5 | |
| HiZOO (LoRA)Optimization Algorithm=HiZOO, Tuning Strategy=LoRA, Model=OPT-13B2026.01 | 63.5 | |
| MeZO (FT)Optimization Algorithm=MeZO, Tuning Strategy=Full Tuning, Guiding Vectors=false, Model=OPT-13B2026.01 | 63.5 | |
| MeZO-GV (Prefix)Optimization Algorithm=MeZO, Tuning Strategy=Prefix Tuning, Guiding Vectors=true, Model=OPT-13B2026.01 | 63.5 | |
| FTOptimization Algorithm=Adam, Tuning Strategy=Full Tuning, Model=OPT-13B2026.01 | 63.5 | |
| Full Fine-Tuning (FT)Model=LLAMA2-7B, Trainable Parameters=6.7B2025.12 | 63.5 | |
| MeZO FTModel Backbone=Llama2-7B, Fine-tuning Strategy=Full Fine-tuning, Optimization Algorithm=MeZO, Number of training examples=10002026.06 | 63.46 | |
| MeZO LoRAModel Backbone=Llama2-7B, Fine-tuning Strategy=LoRA, Optimization Algorithm=MeZO, Number of training examples=10002026.06 | 63.46 | |
| MeZO-GV(LoRA)Backbone=OPT-1.3B, Optimization Method=MeZO, Adaptation Technique=LoRA, Guiding Vectors (GV)=true2026.01 | 62.5 | |
| LoRAModel=LLAMA2-7B, Trainable Parameters=4.2M2025.12 | 62.5 | |
| Dominant-layer ZO LoRAModel Backbone=Llama2-7B, Fine-tuning Strategy=LoRA, Optimization Algorithm=Dominant-layer ZO, Number of training examples=10002026.06 | 62.5 | |
| SubZero (Prefix)Optimization Algorithm=SubZero, Tuning Strategy=Prefix Tuning, Guiding Vectors=false, Model=OPT-13B2026.01 | 62.1 | |
| ZO-AdaMU (2x)Optimization Algorithm=ZO-AdaMU, Tuning Strategy=Full Tuning, Training Steps=2x, Model=OPT-13B2026.01 | 61.5 | |
| Hybrid-PromptModel=Llama-2-7b2026.04 | 61.5 | |
| FO-PrefixModel=Vicuna-7b-v1.52026.04 | 61.5 | |
| FO-PrefixModel=OPT-1.3b2026.04 | 61.5 | |
| Hybrid-LoRAModel=OPT-1.3b2026.04 | 61.5 | |
| MeZO-GV(Prefix)Backbone=OPT-1.3B, Optimization Method=MeZO, Adaptation Technique=Prefix Tuning, Guiding Vectors (GV)=true2026.01 | 60.8 | |
| MeZO-GV(FT)Backbone=OPT-1.3B, Optimization Method=MeZO, Adaptation Technique=Fine-tuning, Guiding Vectors (GV)=true2026.01 | 60.6 | |
| MeZO(Prefix)Backbone=OPT-1.3B, Optimization Method=MeZO, Adaptation Technique=Prefix Tuning2026.01 | 60.6 | |
| HiZOO (Prefix)Optimization Algorithm=HiZOO, Tuning Strategy=Prefix Tuning, Model=OPT-13B2026.01 | 60.6 | |
| MeZO (Prefix)Optimization Algorithm=MeZO, Tuning Strategy=Prefix Tuning, Guiding Vectors=false, Model=OPT-13B2026.01 | 60.6 | |
| FO-LoRAModel=Llama-2-7b2026.04 | 60.6 | |
| Hybrid-LoRAModel=Llama-2-7b2026.04 | 60.6 | |
| ZO-AdaMU (LoRA)Optimization Algorithm=ZO-AdaMU, Tuning Strategy=LoRA, Model=OPT-13B2026.01 | 60.1 | |
| FO-LoRAModel=OPT-1.3b2026.04 | 59.6 | |
| MeZO(LoRA)Backbone=OPT-1.3B, Optimization Method=MeZO, Adaptation Technique=LoRA2026.01 | 58.7 | |
| Hybrid-PromptModel=OPT-1.3b2026.04 | 57.7 | |
| MeZO(FT)Backbone=OPT-1.3B, Optimization Method=MeZO, Adaptation Technique=Fine-tuning2026.01 | 56.7 | |
| ZO-AdaMU (Prefix)Optimization Algorithm=ZO-AdaMU, Tuning Strategy=Prefix Tuning, Model=OPT-13B2026.01 | 56.5 | |
| Hybrid-PromptModel=Vicuna-7b-v1.52026.04 | 55.8 | |
| Hybrid-PrefixModel=OPT-1.3b2026.04 | 55.8 | |
| FO-LoRAModel=Vicuna-7b-v1.52026.04 | 53.8 | |
| FO-PrefixModel=Llama-2-7b2026.04 | 51.9 | |
| ICLBackbone=OPT-1.3B, Learning Strategy=In-context Learning2026.01 | 46.2 | |
| FO-PromptModel=OPT-1.3b2026.04 | 44.2 | |
| Zero-shotBackbone=OPT-1.3B, Learning Strategy=Zero-shot2026.01 | 43.3 | |
| Hybrid-PrefixModel=Llama-2-7b2026.04 | 42.3 | |
| ICLEvaluation Protocol=In-context Learning, Model=OPT-13B2026.01 | 39.4 | |
| Zero-shotEvaluation Protocol=Zero-shot, Model=OPT-13B2026.01 | 38.5 | |
| Zero-shot w/o finetuneModel Backbone=Llama2-7B, Fine-tuning Strategy=None, Optimization Algorithm=None, Number of training examples=10002026.06 | 36.54 | |
| FO-PromptModel=Llama-2-7b2026.04 | 36.5 |