List your company
ResearchFormerly WizWand

Long-form generation factuality and uncertainty estimation on WildHallu (test)

0.86Factuality ScoreLOGU-DPO
0.70920.748350.78750.82665Oct 18, 2024
Updated 8mo ago

Evaluation results

MethodLinks
LOGU-DPOBackbone=Llama3-8B-Instruct, Method Category=Training-Based2024.100.8652.82.68
LOGU-DPOBackbone=Mistral-7B-Instruct, Method Category=Training-Based2024.100.84461.83.49
Pair-FewBackbone=Llama3-8B-Instruct, Method Category=Prompt-Based2024.100.8448.12.33
Unc-FewBackbone=Llama3-8B-Instruct, Method Category=Prompt-Based2024.100.80248.63.44
LOGU-SFTBackbone=Mistral-7B-Instruct, Method Category=Training-Based2024.100.79251.15.73
LOGU-SFTBackbone=Llama3-8B-Instruct, Method Category=Training-Based2024.100.78944.65.37
Unc-ZeroBackbone=Llama3-8B-Instruct, Method Category=Prompt-Based2024.100.78745.83.29
Self-RefineBackbone=Llama3-8B-Instruct, Method Category=Prompt-Based2024.100.7728.75.13
Unc-ZeroBackbone=Mistral-7B-Instruct, Method Category=Prompt-Based2024.100.75448.65.2
Orig.Backbone=Llama3-8B-Instruct, Method Category=Original Model2024.100.744—6.24
Pair-FewBackbone=Mistral-7B-Instruct, Method Category=Prompt-Based2024.100.73451.69.8
Unc-FewBackbone=Mistral-7B-Instruct, Method Category=Prompt-Based2024.100.72657.310.1
Self-RefineBackbone=Mistral-7B-Instruct, Method Category=Prompt-Based2024.100.72450.68.29
Orig.Backbone=Mistral-7B-Instruct, Method Category=Original Model2024.100.715—8.31