List your company

Question Answering on TriviaQA (BLEU, ROUGE-L, METEOR, BS(%))

92.42BS (%)KA2L (10k Unknown)
6.235228.610150.98573.3599Mar 18, 2026Mar 29, 2026Apr 10, 2026Apr 22, 2026May 4, 2026May 16, 2026May 28, 2026

Evaluation results

MethodLinks
KA2L (10k Unknown)Model=Mistral-7B-Instruct-v0.12026.0392.4232.7763.6447.43
KA2L (10k Unknown)Model=Llama-3.1-8B-Instruct2026.0392.3930.8463.3847.05
KA2L (10k Unknown)Model=Mistral-7B-Instruct-v0.32026.0392.2136.8462.4146.66
KA2L (10k Unknown)Model=Llama-2-7b-chat-hf2026.0391.5833.2558.9144.77
KA2L (10k Unknown)Model=glm4-9b-chat2026.039111.6855.842.43
SFT (10k Augmented)Model=Llama-3.1-8B-Instruct2026.0390.1222.0549.5636.23
KA2L (10k Unknown)Model=Qwen1.5-7B-Chat2026.0390.0325.9148.3835.8
KA2L (10k Unknown)Model=Phi-3.5-mini-instruct2026.0389.7325.7247.1534.19
SFT (10k Combine)Model=Llama-3.1-8B-Instruct2026.0389.6523.5949.7836.37
KA2L (5k Unknown)Model=Llama-3.1-8B-Instruct2026.0389.6118.9649.2335.95
SFT (10k Combine)Model=Llama-2-7b-chat-hf2026.0389.3726.3649.1636.84
KA2L (5k Unknown)Model=Llama-2-7b-chat-hf2026.0389.3324.5748.936.69
SFT (10k Augmented)Model=Llama-2-7b-chat-hf2026.0389.1421.9747.5735.75
SFT (10k Combine)Model=Mistral-7B-Instruct-v0.32026.0389.1222.7547.6534.81
SFT (10k Combine)Model=Mistral-7B-Instruct-v0.12026.0389.0818.4146.5133.61
KA2L (5k Unknown)Model=Mistral-7B-Instruct-v0.32026.0389.0221.2147.2134.42
SFT (10k Augmented)Model=Mistral-7B-Instruct-v0.12026.0389.0115.6745.8633.21
SFT (10k Combine)Model=Phi-3.5-mini-instruct2026.0389.0122.9343.3931.6
SFT (10k Augmented)Model=Mistral-7B-Instruct-v0.32026.0388.9916.6347.1334.27
SFT (10k Augmented)Model=Phi-3.5-mini-instruct2026.0388.9919.2642.9831.27
KA2L (5k Unknown)Model=Phi-3.5-mini-instruct2026.0388.9823.2642.9431.24
KA2L (5k Unknown)Model=Mistral-7B-Instruct-v0.12026.0388.9215.9345.5232.98
SFT (5k Known)Model=Llama-3.1-8B-Instruct2026.0388.514.8143.1431.54
SFT (5k Known)Model=Qwen1.5-7B-Chat2026.0388.420.0439.0528.53
KA2L (5k Unknown)Model=Qwen1.5-7B-Chat2026.0388.419.8939.0228.54
SFT (10k Combine)Model=Qwen1.5-7B-Chat2026.0388.3920.7838.8528.16
SFT (5k Known)Model=Phi-3.5-mini-instruct2026.0388.2918.1339.128.31
SFT (10k Augmented)Model=Qwen1.5-7B-Chat2026.0388.2413.8538.0627.56
SFT (10k Combine)Model=glm4-9b-chat2026.0388.1821.5141.2630.71
SFT (10k Augmented)Model=glm4-9b-chat2026.0388.0915.2140.4730.07
SFT (5k Known)Model=Llama-2-7b-chat-hf2026.0388.0619.6741.8631.33
KA2L (5k Unknown)Model=glm4-9b-chat2026.0387.9719.6240.2829.85
SFT (5k Known)Model=Mistral-7B-Instruct-v0.32026.0387.9416.8740.8729.36
SFT (5k Known)Model=Mistral-7B-Instruct-v0.12026.0387.8813.239.3528.41
KA2L (10k Unknown)Model=DeepSeek-R1-Distill-Qwen-7B2026.0387.5614.22718.86
SFT (10k Augmented)Model=DeepSeek-R1-Distill-Qwen-7B2026.0387.166.6824.4617.11
SFT (10k Combine)Model=DeepSeek-R1-Distill-Qwen-7B2026.0387.059.9823.6516.38
KA2L (5k Unknown)Model=DeepSeek-R1-Distill-Qwen-7B2026.0386.859.4122.2215.39
KA2L (10k Unknown)Model=Qwen2.5-7B-Instruct2026.0386.4914.4935.8827.07
SFT (5k Known)Model=glm4-9b-chat2026.0386.3414.6831.6523.03
SFT (5k Known)Model=DeepSeek-R1-Distill-Qwen-7B2026.0386.327.7918.7712.82
SFT (10k Augmented)Model=Qwen2.5-7B-Instruct2026.0385.85.8231.9924.24
SFT (10k Combine)Model=Qwen2.5-7B-Instruct2026.0385.6511.3732.5224.36
SFT (5k Known)Model=Qwen2.5-7B-Instruct2026.0385.6311.7232.2424.18
KA2L (5k Unknown)Model=Qwen2.5-7B-Instruct2026.0385.6311.7232.324.2
None (Zero-shot)Model=Mistral-7B-Instruct-v0.12026.0381.850.3215.7317.64
None (Zero-shot)Model=Qwen1.5-7B-Chat2026.0379.280.143.848.18
None (Zero-shot)Model=Llama-3.1-8B-Instruct2026.0379.260.125.6711.11
None (Zero-shot)Model=glm4-9b-chat2026.0379.130.153.628.16
None (Zero-shot)Model=Mistral-7B-Instruct-v0.32026.0379.070.163.147.11
None (Zero-shot)Model=Qwen2.5-7B-Instruct2026.0378.920.154.078.79
None (Zero-shot)Model=Phi-3.5-mini-instruct2026.0378.860.112.415.85
None (Zero-shot)Model=Llama-2-7b-chat-hf2026.0378.230.092.35.24
None (Zero-shot)Model=DeepSeek-R1-Distill-Qwen-7B2026.0376.130.020.491.24
+ Scaling-bin.Calibration category=Post-hoc plurality calibrators2026.0512.29———
GraphCalCalibration category=Trained calibrators2026.0512.29———
Plurality shareCalibration category=Post-hoc plurality calibrators2026.0512.12———
Scalar + GBTCalibration category=Trained calibrators2026.0511.43———
+ PlattCalibration category=Post-hoc plurality calibrators2026.0510.91———
+ IsotonicCalibration category=Post-hoc plurality calibrators2026.0510.89———
LLM-Cal (+topo)Calibration category=LLM-elicited confidence estimators2026.0510.75———
DiscoUQ-LLMCalibration category=Trained calibrators2026.0510.47———
Collab. Cal.Calibration category=LLM-elicited confidence estimators2026.0510.32———
LLM-Cal (no topo)Calibration category=LLM-elicited confidence estimators2026.0510.11———
CAGE-CALCalibration category=Trained calibrators2026.059.55———