List your company
ResearchFormerly WizWand

Hallucination robustness on TruthfulQA

48.76TruthfulQA AccuracyQwen3-1.7B
47.896848.120948.34548.5691May 28, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Qwen3-1.7BType=Base Model2026.0548.76
FoLoRAFine-tuning dataset=MetaMathQA, Calibration data=NQ Open, Calibration model size=4B2026.0548.46
LoRA-NullFine-tuning dataset=MetaMathQA, Calibration data=model-generated pretraining-proxy2026.0548.4
FoLoRAFine-tuning dataset=MetaMathQA, Calibration data=NQ Open, Calibration model size=1.7B2026.0548.36
CorDAFine-tuning dataset=MetaMathQA, Calibration data=model-generated pretraining-proxy, Calibration model size=4B2026.0548.35
FoLoRAFine-tuning dataset=MetaMathQA, Calibration data=NQ Open2026.0548.32
CorDAFine-tuning dataset=MetaMathQA, Calibration data=model-generated pretraining-proxy2026.0548.3
FoLoRAFine-tuning dataset=MetaMathQA, Calibration data=NQ Open, Calibration model size=0.6B2026.0548.22
LoRA-NullFine-tuning dataset=MetaMathQA, Calibration data=model-generated pretraining-proxy, Calibration model size=4B2026.0548.16
MiLoRAFine-tuning dataset=MetaMathQA2026.0548.06
OPLoRAFine-tuning dataset=MetaMathQA2026.0548.06
LoRAFine-tuning dataset=MetaMathQA2026.0547.93