List your company

Retrieval-Augmented Question Answering on NQ, TriviaQA, and WebQ (test)

62.2Answer Quality RecallERA
-1.03215.38431.848.216Feb 24, 2026

Evaluation results

MethodLinks
ERABackbone=Qwen3-8B-Base2026.0262.253.257.360.645.440.871.55254.5
ERABackbone=Llama3.1-8B2026.0253.7686050.436.775.950.660.760.4
LogprobBackbone=Llama3.1-8B, t=-22026.0250.99165.356.839.231.442.436.146.5
DTABackbone=Llama3.1-8B2026.024954.451.65132.259.648.153.352.4
ICLBackbone=Llama3.1-8B2026.0247.433.639.454.832.61.4802.75
DTABackbone=Qwen3-8B-Base2026.0243.64142.23831.947.656.651.746.5
Self-ConsistencyBackbone=Llama3.1-8B2026.0238.649.243.239.23055.936.844.443.8
ptrueBackbone=Llama3.1-8B, t=0.52026.0236.493.152.337.523.531.155.626.831.2
ptrueBackbone=Llama3.1-8B, t=0.72026.0233.793.449.535.220.749.748.429.336.8
LogprobBackbone=Qwen3-8B-Base, t=-22026.0230.893.346.330.322.1556.49.115.3
LogprobBackbone=Llama3.1-8B, t=-12026.0228.291.443.134.619.977.53447.245.1
ptrueBackbone=Llama3.1-8B, t=0.92026.0225.395.240301377.940.222.225.5
ptrueBackbone=Qwen3-8B-Base, t=0.52026.0220.993.134.224.512.852.139.945.238.9
Self-ConsistencyBackbone=Qwen3-8B-Base2026.0220.341.527.321.914.277.73548.334.9
LogprobBackbone=Qwen3-8B-Base, t=-12026.0219.793.332.518.714.468.93647.338.5
ICLBackbone=Qwen3-8B-Base2026.0217.8131518.414.24.840.48.510.9
ptrueBackbone=Qwen3-8B-Base, t=0.72026.0213.397.323.416.13.688.335.350.432
ptrueBackbone=Qwen3-8B-Base, t=0.92026.021.41002.91.40.19929.445.35.4