List your company

Question Answering Calibration on OOD Evaluation (Squad, WQ, CWQ, MSQ, PopQA)

5Squad Calibration ScoreThermometer
3.3614.4325.536.57Oct 20, 2025

Evaluation results

MethodLinks
ThermometerBackbone=Qwen2.5-7B-Instruct2025.105516107
DACABackbone=Qwen2.5-7B-Instruct2025.1098661511
EliCal(1k)Backbone=Qwen2.5-7B-Instruct, LoRA rank=8, LoRA alpha=16, Batch size=162025.101065826
Consis-LexBackbone=Qwen2.5-7B-Instruct2025.1014713322620
Cal-Only(1k)Backbone=Qwen2.5-7B-Instruct, LoRA rank=8, LoRA alpha=16, Batch size=162025.1017151612712
ProbBackbone=Qwen2.5-7B-Instruct2025.10315433131625
Consis-SemBackbone=Qwen2.5-7B-Instruct2025.10311516212927
ElicitationBackbone=Qwen2.5-7B-Instruct, LoRA rank=8, LoRA alpha=16, Batch size=162025.1031911192424
Verbal-10Backbone=Qwen2.5-7B-Instruct2025.10353236463636
Verbal-0Backbone=Qwen2.5-7B-Instruct2025.10422944465748
N-ProbBackbone=Qwen2.5-7B-Instruct2025.10462446656052