List your company
ResearchFormerly WizWand

Accuracy on Natural Language Inference on CB

98.2AccuracyFADS-ICL
55.497666.583877.6788.7562Dec 2, 2022Jun 16, 2023Dec 30, 2023Jul 13, 2024Jan 26, 2025Aug 10, 2025Feb 23, 2026
Updated 5mo ago

Evaluation results

MethodLinks
FADS-ICLBackbone=Llama-2 70B, Shots=128-shots2024.0598.2
FADS-ICLBackbone=Llama-2 13B, Shots=128-shots2024.0596.1
Individual2024.0595.8
Traditional MTL2024.0595.8
ICLBackbone=Llama-2 70B, Shots=128-shots2024.0595.4
FADS-ICLBackbone=Llama-1 7B, Shots=128-shots2024.0595
kNN-promptingBackbone=Llama-2 70B, Shots=128-shots2024.0594.3
FADS-ICLBackbone=Llama-1 30B, Shots=128-shots2024.0592.6
FADS-ICLBackbone=Llama-1 13B, Shots=128-shots2024.0592.3
kNN-promptingBackbone=Llama-2 13B, Shots=128-shots2024.0591.4
FADS-ICLBackbone=Llama-2 7B, Shots=128-shots2024.0590.7
Falcon-180Bprotocol=1-shot2023.1189.3
kNN-promptingBackbone=Llama-1 13B, Shots=128-shots2024.0588.9
kNN-promptingBackbone=Llama-1 30B, Shots=128-shots2024.0588.9
kNN-promptBackbone=Llama-2 70B, Shots=128-shots2024.0587.9
PaLM 2-Lprompting=1-shot2023.0587.5
PaLM-2 Lprotocol=1-shot2023.1187.5
Ties-MergingValidation=false2024.0587.5
EMR-MERGINGValidation=false2024.0587.5
kNN-promptBackbone=Llama-2 13B, Shots=128-shots2024.0586.1
FADS-ICLNumber of training shots (m)=32, LLM scale=1.5B2024.0585.9
ColD-Fusion2022.1285
kNN-promptingBackbone=Llama-1 7B, Shots=128-shots2024.0584.3
UCPOFModel=Llama-3-8B-Instruct2026.0283.93
gold shot2026.0283.93
PaLMprompting=1-shot2023.0583.9
PaLMprotocol=1-shot2023.1183.9
Fisher MergingValidation=true2024.0583.3
Task ArithmeticValidation=true2024.0583.3
Ties-MergingValidation=true2024.0583.3
FADS-ICLNumber of training shots (m)=128, LLM scale=1.5B2024.0583.2
FADS-ICLBackbone=GPT-2 1.5B, Shots=128-shots2024.0583.2
FADS-ICLcandidate pool size (m)=1282024.0583.2
FADS-ICLcandidate pool size (m)=2562024.0583.2
Multitask2022.1282.86
Full RAGModel=Llama-3-8B-Instruct2026.0282.14
UCPOF2026.0282.14
PaLM 2-Sprompting=1-shot2023.0582.1
PaLM-2 Sprotocol=1-shot2023.1182.1
PaLM 2-Mprompting=1-shot2023.0580.4
PaLM-2 Mprotocol=1-shot2023.1180.4
kNN-promptBackbone=Llama-1 7B, Shots=128-shots2024.0580.4
MUPPET2022.1280.36
FADS-ICLBackbone=GPT-2 0.8B, Shots=128-shots2024.0580
kNN-promptBackbone=Llama-2 7B, Shots=128-shots2024.0580
kNN-promptBackbone=Llama-1 13B, Shots=128-shots2024.0580
ICLBackbone=Llama-1 30B, Shots=128-shots2024.0580
kNN-promptBackbone=Llama-1 30B, Shots=128-shots2024.0580
Task ArithmeticValidation=false2024.0579.2
FADS-ICLNumber of training shots (m)=64, LLM scale=1.5B2024.0578.6
BERT Large FTBackbone=BERT Large, Shots=N/A (Fine-tuned)2024.0578.6
FADS-ICLcandidate pool size (m)=642024.0578.6
baseline2026.0278.57
kNN-promptingBackbone=Llama-2 7B, Shots=128-shots2024.0577.1
FO (Adam)Base Model=LLaMA-3.2-1B, Optimization Regime=First-Order2025.1075
ICLBackbone=Llama-1 7B, Shots=128-shots2024.0574.6
SimCSEcandidate pool size (m)=1282024.0573.2
Trans-Encodercandidate pool size (m)=1282024.0573.2
SimCSEcandidate pool size (m)=2562024.0573.2
Trans-Encodercandidate pool size (m)=2562024.0573.2
ZO Fine-tunerBase Model=LLaMA-3.2-1B, Optimization Regime=Zeroth-Order2025.1073
FADS-ICLcandidate pool size (m)=322024.0572.1
kNN-promptingNumber of training shots (m)=32, LLM scale=1.5B2024.0571.8
SimCSEcandidate pool size (m)=642024.0571.4
BM25candidate pool size (m)=1282024.0571.4
BM25candidate pool size (m)=2562024.0571.4
GPT Large FTBackbone=GPT Large, Shots=N/A (Fine-tuned)2024.0570
ICLBackbone=Llama-2 7B, Shots=128-shots2024.0570
MeZOBase Model=LLaMA-3.2-1B, Optimization Regime=Zeroth-Order2025.1070
SBERTcandidate pool size (m)=1282024.0569.6
SBERTcandidate pool size (m)=2562024.0569.6
Trans-Encodercandidate pool size (m)=642024.0569.3
BM25candidate pool size (m)=642024.0568.6
kNN-promptingBackbone=GPT-2 0.8B, Shots=128-shots2024.0567.5
SBERTcandidate pool size (m)=642024.0566.8
SimCSEcandidate pool size (m)=322024.0566.4
kNN-promptingNumber of training shots (m)=128, LLM scale=1.5B2024.0565
kNN-promptingBackbone=GPT-2 1.5B, Shots=128-shots2024.0565
BM25candidate pool size (m)=322024.0565
GPT-3Evaluation Protocol=1-shot2023.0364.3
Finetune2022.1264.29
baselineModel=Llama-3-8B-Instruct2026.0264.29
FADS-ICLNumber of training shots (m)=16, LLM scale=1.5B2024.0563.9
SBERTcandidate pool size (m)=322024.0562.9
ICLBackbone=Llama-1 13B, Shots=128-shots2024.0562.5
ICLBackbone=Llama-2 13B, Shots=128-shots2024.0562.5
gold shotModel=Llama-3-8B-Instruct2026.0262.5
kNN-promptingNumber of training shots (m)=64, LLM scale=1.5B2024.0561.4
Trans-Encodercandidate pool size (m)=322024.0561.4
In-Context Learning2024.0560.7
RegMeanValidation=true2024.0558.3
Weight AveragingValidation=false2024.0558.3
ICLNumber of training shots (m)=4, LLM scale=1.5B2024.0557.9
ICLNumber of training shots (m)=8, LLM scale=1.5B2024.0557.9
ICLNumber of training shots (m)=16, LLM scale=1.5B2024.0557.9
ICLNumber of training shots (m)=32, LLM scale=1.5B2024.0557.9
ICLNumber of training shots (m)=64, LLM scale=1.5B2024.0557.9
ICLNumber of training shots (m)=128, LLM scale=1.5B2024.0557.9
ICLBackbone=GPT-2 1.5B, Shots=128-shots2024.0557.9
MaskProBase Model=LLaMA-2-7B, Sparsity=2:42025.0657.14