List your company
ResearchFormerly WizWand

Instruction following on MT-Bench GPT-4 scored (test)

6.31ScoreVB-LoRA
4.72925.13965.555.9604May 24, 2024Sep 23, 2024Jan 23, 2025May 26, 2025Sep 25, 2025Jan 25, 2026May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
VB-LoRAModel=Llama2-13B, # Parameters=1.1M2024.056.31
FoRAParams (M)=80.0, Backbone=LLaMA-2-7B, Fine-tuning dataset=Alpaca-52K2026.056.15
LoRA#Model=Llama2-13B, # Parameters=250.3M2024.056.13
DoRAParams (M)=160.5, Backbone=LLaMA-2-7B, Fine-tuning dataset=Alpaca-52K2026.056
VeRAModel=Llama2-13B, # Parameters=2.4M2024.055.93
rsLoRAParams (M)=159.9, Backbone=LLaMA-2-7B, Fine-tuning dataset=Alpaca-52K2026.055.82
LoRAModel=Llama2-13B, # Parameters=250.3M2024.055.77
VB-LoRAModel=Llama2-7B, # Parameters=0.8M2024.055.71
LoRAParams (M)=159.9, Backbone=LLaMA-2-7B, Fine-tuning dataset=Alpaca-52K2026.055.7
LoRA#Model=Llama2-7B, # Parameters=159.9M2024.055.63
w/o FTModel=Llama2-13B2024.055.38
LoRA†Model=Llama2-7B, # Parameters=159.9M2024.055.19
VeRAModel=Llama2-7B, # Parameters=1.6M2024.055.08
w/o FTModel=Llama2-7B2024.054.79