List your company
ResearchFormerly WizWand

Multiple Choice on COPA

100AccuracyFull Fine-Tuning (FT)
62.45672.20381.9591.697Dec 27, 2025Jan 22, 2026Feb 17, 2026Mar 16, 2026Apr 11, 2026May 7, 2026Jun 3, 2026
Updated 4mo ago

Evaluation results

MethodLinks
Full Fine-Tuning (FT)Model=Qwen2.5-14B, Trainable Parameters=14.7B2025.12100
LoRAModel=Qwen2.5-14B, Trainable Parameters=6.3M2025.12100
Partial-LoRAModel=Qwen2.5-14B, Trainable Parameters=122K2025.12100
Masking (0.001%)Model=Qwen2.5-14B, Trainable Parameters=149K2025.1298
Partial-LoRAModel=LLAMA2-7B, Trainable Parameters=80K2025.1290
Full Fine-Tuning (FT)Model=LLAMA3.1-8B, Trainable Parameters=8B2025.1290
LoRAModel=LLAMA3.1-8B, Trainable Parameters=3.4M2025.1289
Masking (0.001%)Model=LLAMA2-7B, Trainable Parameters=68K2025.1288
Partial-LoRAModel=LLAMA3.1-8B, Trainable Parameters=92k2025.1288
AR1-ZOModel=Qwen3-32B, Rank (r)=642026.0588
MeZO LoRAModel Backbone=Llama2-7B, Fine-tuning Strategy=LoRA, Optimization Algorithm=MeZO, Number of training examples=10002026.0688
Full Fine-Tuning (FT)Model=LLAMA2-7B, Trainable Parameters=6.7B2025.1287
Dominant-layer ZO FTModel Backbone=Llama2-7B, Fine-tuning Strategy=Full Fine-tuning, Optimization Algorithm=Dominant-layer ZO, Number of training examples=10002026.0687
Dominant-layer ZO LoRAModel Backbone=Llama2-7B, Fine-tuning Strategy=LoRA, Optimization Algorithm=Dominant-layer ZO, Number of training examples=10002026.0687
LOZOModel=OPT-13B2026.0586
AR1-ZOModel=OPT-13B, Rank (r)=642026.0586
First Order Adamw FTModel Backbone=Llama2-7B, Fine-tuning Strategy=Full Fine-tuning, Optimization Algorithm=First Order Adamw, Number of training examples=10002026.0686
MeZO FTModel Backbone=Llama2-7B, Fine-tuning Strategy=Full Fine-tuning, Optimization Algorithm=MeZO, Number of training examples=10002026.0686
LoRAModel=LLAMA2-7B, Trainable Parameters=4.2M2025.1285
MeZO-LoRAModel=OPT-13B, Rank (r)=642026.0584
FT (Adam)Model=OPT-2.7B2026.0581
Zero-shot w/o finetuneModel Backbone=Llama2-7B, Fine-tuning Strategy=None, Optimization Algorithm=None, Number of training examples=10002026.0681
LOZOModel=OPT-2.7B2026.0580
AR1-ZOModel=OPT-2.7B, Rank (r)=642026.0579
FT (Adam)Model=OPT-13B2026.0579
CAQ-ZOBackbone=Llama-2-7B, Precision=NF42026.0578.4
MeZOBackbone=Llama-2-7B, Precision=BF162026.0578
Zero-Shot-QBackbone=Llama-2-7B, Precision=NF42026.0578
QuZOBackbone=Llama-2-7B, Precision=NF42026.0577.9
MeZO-LoRAModel=Qwen3-32B, Rank (r)=642026.0577
CAQ-ZOBackbone=Qwen-2.5-1.5B, Precision=NF42026.0576.2
Masking (0.001%)Model=LLAMA3.1-8B, Trainable Parameters=80K2025.1276
MeZOBackbone=Qwen-2.5-1.5B, Precision=BF162026.0575.1
ZO-Alt-NaiveModel=OPT-13B2026.0575
Zero-Shot-QBackbone=Qwen-2.5-1.5B, Precision=NF42026.0573
ZO-Alt-NaiveModel=Qwen3-32B2026.0573
MeZO-LoRAModel=OPT-2.7B, Rank (r)=642026.0572
AR1-ZOModel=Qwen3-1.7B, Rank (r)=642026.0572
ZO-Alt-NaiveModel=OPT-2.7B2026.0571
MeZO-LoRAModel=Qwen3-1.7B, Rank (r)=642026.0571
ZO-Alt-NaiveModel=Qwen3-1.7B2026.0571
QuZOBackbone=Qwen-2.5-1.5B, Precision=NF42026.0563.9