List your company
ResearchFormerly WizWand

Commonsense Reasoning on BoolQ (ACC, ECE, NLL)

87.6AccuracyBayesLoRA
54.663263.214171.76580.3159Jun 17, 2024Oct 8, 2024Jan 29, 2025May 22, 2025Sep 12, 2025Jan 3, 2026Apr 27, 2026
Updated 5mo ago

Evaluation results

MethodLinks
BayesLoRAk=10, Params (M)=2.40–3.872025.0687.66.870.38
BayesLoRAk=0, Params (M)=2.40–3.872025.0687.549.730.49
BayesLoRAr=8, k=0, Params (M)=4.482025.0687.4210.110.5
MCDBackbone=Llama2-7B, Method=LoRA, Steps=5,0002024.0687.293.460.31
MCDParams (M)=4.482025.0687.293.460.31
BBBBackbone=Llama2-7B, Method=LoRA, Steps=5,0002024.0687.213.740.31
BBBParams (M)=9.982025.0687.213.740.31
BayesLoRAr=8, k=10, Params (M)=4.482025.0687.147.340.38
ENSBackbone=Llama2-7B, Method=LoRA, Steps=5,0002024.0687.099.610.57
ENSParams (M)=44.802025.0687.099.610.57
BLoBBackbone=Llama2-7B, Method=LoRA, Steps=5,000, N=102024.0686.991.410.31
BLoB (N=10)N (Ensemble Size)=10, rind=9, cind=92026.0186.991.410.31
LAPBackbone=Llama2-7B, Method=LoRA, Steps=5,0002024.0686.951.30.31
BLoBBackbone=Llama2-7B, Method=LoRA, Steps=5,000, N=02024.0686.82.460.32
MLEBackbone=Llama2-7B, Method=LoRA, Steps=5,0002024.0686.533.180.32
LoRAParams (M)=4.482025.0686.533.180.32
MAPBackbone=Llama2-7B, Method=LoRA, Steps=5,0002024.0686.53.160.33
BLoBBackbone=Llama2-7B, Method=LoRA, Steps=5,000, N=52024.0686.471.630.32
Bayesian-LoRA (S=4)S (Samples)=4, Evaluation Protocol=End-to-End, rind=9, cind=92026.0186.12.10.29
Dropoutrind=9, cind=92026.0185.97.50.43
MAPrind=9, cind=92026.0185.87.40.43
Temprind=9, cind=92026.0185.82.50.35
LLLA (post-hoc)Evaluation Protocol=post-hoc, rind=9, cind=92026.0185.81.70.35
LA (post-hoc)Evaluation Protocol=post-hoc, rind=9, cind=92026.0185.81.90.34
Ckpt Ensrind=9, cind=92026.0185.43.20.35
BBBrind=9, cind=92026.0183.173.740.31
BaLoRAModel=Llama-3-8B, Params (%)=0.71452026.0476.42——
HiRAModel=Llama-3-8B, Params (%)=0.70022026.0475.4——
DoRAModel=Llama-3-8B, Params (%)=0.70022026.0474.6——
MoRAModel=Llama-3-8B, Params (%)=0.69972026.0474.28——
ChatGPTModel=ChatGPT2026.0473.1——
BaLoRAModel=Llama-2-7B, Params (%)=0.84272026.0472.69——
MoRAModel=Llama-2-7B, Params (%)=0.82412026.0472.17——
DoRAModel=Llama-2-7B, Params (%)=0.82562026.0471.8——
HiRAModel=Llama-2-7B, Params (%)=0.82562026.0471.22——
LoRAModel=Llama-3-8B, Params (%)=0.70022026.0470.8——
LoRAModel=Llama-2-7B, Params (%)=0.82562026.0469.8——
P-TuningModel=Llama-3-8B, Params (%)=0.62402026.0459.97——
P-TuningModel=Llama-2-7B, Params (%)=0.74282026.0458.75——
Prompt TuningModel=Llama-3-8B, Params (%)=0.00102026.0456.85——
Prompt TuningModel=Llama-2-7B, Params (%)=0.00122026.0455.93——