List your company
ResearchFormerly WizWand

Natural Language Inference on ANLI Round 3

67.9AccuracyLMSI
29.752839.656449.5659.4636Apr 14, 2022Aug 20, 2022Dec 26, 2022May 4, 2023Sep 9, 2023Jan 15, 2024May 23, 2024
Updated 8mo ago

Evaluation results

MethodLinks
LMSIPrompting Method=Self-Consistency2022.1067.9
LMSIPrompting Method=CoT-Prompting2022.1067.3
LMSIPrompting Method=Standard-Prompting2022.1066.9
Wang et al. (2022a)description=Previous SOTA2022.1066
Self-ConsistencyLMSI=false2022.1063.4
CoT-PromptingLMSI=false2022.1060.6
FLAN-T5Model Variant=xlarge, Model Parameters=3B2023.0756.6
Standard-PromptingLMSI=false2022.1055.8
Individual2024.0553
ALIGNModel Variant=large, Model Parameters=355M2023.0752.3
Ties-MergingValidation=true2024.0551.1
EMR-MERGINGValidation=false2024.0550.8
Task ArithmeticValidation=true2024.0550
FLAN-T5Model Variant=large, Model Parameters=780M2023.0749.8
Task ArithmeticValidation=false2024.0548.2
Traditional MTL2024.0547.7
Ties-MergingValidation=false2024.0547.4
FLAN 137BEvaluation protocol=0-shot, Training strategy=leave-one-category-out2022.1247
ALIGNModel Variant=base, Model Parameters=125M2023.0745.5
OPT-IML 175BEvaluation protocol=few-shot, k=52022.1244.1
OPT-IML 175BEvaluation protocol=0-shot2022.1243.8
FLAN 137BEvaluation protocol=few-shot, k=variable, Training strategy=leave-one-category-out2022.1242.8
Fisher MergingValidation=true2024.0542.2
T5(3B) + PE w/ ROE (ORC.)Backbone=T5 (3B), Expert type=Oracle Retrieval-of-Expert, Additional Parameters=100M2023.0242.07
T0-11BParameters=11B2023.0241.26
LaMDA-PT 137BEvaluation protocol=few-shot, k=variable2022.1240.7
RegMeanValidation=true2024.0540.2
Weight AveragingValidation=false2024.0540.2
OPT-IML 30BEvaluation protocol=0-shot2022.1239.6
LaMDA-PT 137BEvaluation protocol=0-shot2022.1239.3
OPT-IML 30BEvaluation protocol=few-shot, k=52022.1238.3
BloombergGPTEvaluation Protocol=1-shot2023.0337.33
FairSeqNumber of Parameters=1.3B, Evaluation Protocol=5-shot2022.0437
GPT-3Model Variant=DaVinci, Zero-shot=true2022.0436.9
FairSeqNumber of Parameters=6.7B, Evaluation Protocol=5-shot2022.0436.7
T5(3B) + Cos PEBackbone=T5 (3B), Expert type=Prompt Expert trained on COSMOS-QA, Additional Parameters=100M2023.0236.38
GPT-NeoXEvaluation Protocol=1-shot2023.0336.17
FairSeqNumber of Parameters=125M, Evaluation Protocol=5-shot2022.0435.9
FairSeqNumber of Parameters=13B, Evaluation Protocol=5-shot2022.0435.7
GPT-JModel Size=6B, Zero-shot=true2022.0435.5
GPT-NeoXModel Size=20B, Zero-shot=true2022.0435.4
GPT-3Model Variant=Ada, Zero-shot=true2022.0435.4
GPT-3Model Variant=Curie, Zero-shot=true2022.0435.3
BLOOM-176BEvaluation Protocol=1-shot2023.0335.17
GPT-3Evaluation Protocol=1-shot2023.0335.1
OPT-66BEvaluation Protocol=1-shot2023.0334.92
FairSeqNumber of Parameters=13B, Zero-Shot=true2022.0434.7
FairSeqNumber of Parameters=355M, Evaluation Protocol=5-shot2022.0434.7
GPT-JParameters=6B, Evaluation=Five-shot2022.0434.6
OPT 175BEvaluation protocol=few-shot, k=52022.1234.6
GPT-3 (175B)Parameters=175B2023.0234.5
GPT-NeoXParameters=20B, Evaluation=Five-shot2022.0434.2
FairSeqNumber of Parameters=2.7B, Zero-Shot=true2022.0434
GPT-3Model Variant=Babbage, Zero-shot=true2022.0434
OPT 30BEvaluation protocol=0-shot2022.1233.5
OPT 30BEvaluation protocol=few-shot, k=52022.1233.5
OPT 175BEvaluation protocol=0-shot2022.1233.5
T0-3BParameters=3B2023.0233.32
FairSeqNumber of Parameters=1.3B, Zero-Shot=true2022.0433.3
FairSeqNumber of Parameters=6.7B, Zero-Shot=true2022.0433.3
FairSeqNumber of Parameters=125M, Zero-Shot=true2022.0433
FairSeqNumber of Parameters=2.7B, Evaluation Protocol=5-shot2022.0432.6
FairSeqNumber of Parameters=355M, Zero-Shot=true2022.0432.3
T5(3B) + PE w/ ROEBackbone=T5 (3B), Expert type=Retrieval-of-Expert, Additional Parameters=100M2023.0231.22