List your company
ResearchFormerly WizWand

Question Answering on SQuAD

93.33Exact MatchCortexDebate
11.617232.831154.04575.2589Feb 20, 2024Jun 26, 2024Nov 1, 2024Mar 8, 2025Jul 14, 2025Nov 18, 2025Mar 26, 2026
Updated 7mo ago

Evaluation results

MethodLinks
CortexDebateType=Ours2025.0793.33
GDType=Part Debate2025.0790.33
SFTBackbone=Mistral 7B2026.0389.67
SFTBackbone=Llama3.1 8B2026.0389.49
CoRectModel=Qwen2.5-7b-instruct2026.0288.93
COIECDModel=Qwen2.5-7b-instruct2026.0288.9
RECONCILEType=Full Debate2025.0788.33
NDType=Part Debate2025.0788.33
PRDType=Full Debate2025.0787.67
ChatEvalType=Full Debate2025.0787.33
SFTBackbone=Qwen3 1.7B2026.0386.64
MLDType=Full Debate2025.0786.33
CoABackbone=Llama3.1 8B2026.0386.13
MaVType=No Debate2025.0785.33
CADModel=LLaMA3-8b-instruct2026.0284.93
CoABackbone=Qwen3 1.7B2026.0384.68
CADModel=Qwen2.5-7b-instruct2026.0284.43
COIECDModel=LLaMA3-8b-instruct2026.0284.4
CoRectModel=LLaMA3-8b-instruct2026.0283.7
HyperMoEArchitecture=Switch Transformer2024.0283.51
CoABackbone=Mistral 7B2026.0383.15
MoEArchitecture=Switch Transformer2024.0283.01
MoE-ShareArchitecture=Switch Transformer2024.0282.87
GreedyModel=LLaMA3-8b-instruct2026.0281.93
AdaCADModel=LLaMA3-8b-instruct2026.0281.87
GreedyModel=Qwen2.5-7b-instruct2026.0281.27
AdaCADModel=Qwen2.5-7b-instruct2026.0281.2
T5-baseModel Size=220M, Data=Full, Method=Fine-Tuned2024.0377.3
GOLDBase Model=LLaMA2 -> T5-base, Model Size=220M, Data=3K2024.0375.2
T5-baseModel Size=220M, Data=Full, Method=Pre-Trained2024.0374.7
P2ModelBase Model=LLaMA2 -> T5-base, Model Size=220M, Data=3K2024.0374.4
CoRectModel=Mistral-7b2026.0274.33
COIECDModel=Mistral-7b2026.0274.2
CADModel=Mistral-7b2026.0272.9
CoRectModel=LLaMA2-13b-chat2026.0270.7
ZeroGenBase Model=LLaMA2 -> T5-base, Model Size=220M, Data=3K2024.0369.4
CADModel=LLaMA2-13b-chat2026.0268.97
sudoLMBackbone=Qwen3 1.7B2026.0368.48
ProGenBase Model=LLaMA2 -> T5-base, Model Size=220M, Data=3K2024.0368.1
BEST-ON-VALBackbone=Llama-22024.1266.1
Self-RAGLLM=Llama-3.1-8B, Write-back=true2026.0365.45
FLARELLM=Llama-3.1-8B, Write-back=true2026.0364.16
GreedyModel=Mistral-7b2026.0263.93
AdaCADModel=Mistral-7b2026.0263.87
Self-RAGLLM=Llama-3.1-8B, Write-back=false2026.0363.58
RePlugLLM=Llama-3.1-8B, Write-back=true2026.0363.42
Naive RAGLLM=Llama-3.1-8B, Write-back=true2026.0363.23
Self-RAGLLM=Gemma-3-12B, Write-back=true2026.0363.22
SMOOTHIE-LOCALBackbone=Llama-22024.1263.1
FLARELLM=Llama-3.1-8B, Write-back=false2026.0362.29
Naive RAGLLM=Llama-3.1-8B, Write-back=false2026.0362.14
RePlugLLM=Llama-3.1-8B, Write-back=false2026.0361.83
FLARELLM=Gemma-3-12B, Write-back=true2026.0361.78
Self-RAGLLM=Gemma-3-12B, Write-back=false2026.0361.73
SMOOTHIE-GLOBALBackbone=Llama-22024.1261.4
RePlugLLM=Gemma-3-12B, Write-back=true2026.0361.39
Naive RAGLLM=Gemma-3-12B, Write-back=true2026.0361.24
Naive RAGLLM=Gemma-3-12B, Write-back=false2026.0360.89
FLARELLM=Gemma-3-12B, Write-back=false2026.0360.55
RePlugLLM=Gemma-3-12B, Write-back=false2026.0360.25
COIECDModel=LLaMA2-13b-chat2026.0258.67
No RetrievalLLM=Llama-3.1-8B2026.0355.18
LLaMA2Model Size=7B, Method=Few-shot, Prompt=Specific2024.0354.7
No RetrievalLLM=Gemma-3-12B2026.0354.36
BaseBackbone=Llama3.1 8B2026.0354.19
RANDOMBackbone=Llama-22024.1254
PermLMBackbone=Llama3.1 8B2026.0353.9
SMOOTHIE-GLOBALBackbone=Falcon2024.1253.3
SMOOTHIE-LOCALBackbone=Falcon2024.1253.3
BEST-ON-VALBackbone=Falcon2024.1252.7
GreedyModel=LLaMA2-13b-chat2026.0252.17
AdaCADModel=LLaMA2-13b-chat2026.0251.8
RANDOMBackbone=Falcon2024.1247.3
dual-debiasing frameworkRetriever=DPP, Noise Type=None, Noise Ratio=0.02025.0537.2
dual-debiasing frameworkRetriever=Random, Noise Type=Irrelevant, Noise Ratio=0.82025.0535.2
Naïve ICLRetriever=TopK, Noise Type=Relevant, Noise Ratio=0.22025.0533.47
BaseBackbone=Qwen3 1.7B2026.0331.95
PermLMBackbone=Qwen3 1.7B2026.0331.95
BaseBackbone=Mistral 7B2026.0329.1
PermLMBackbone=Mistral 7B2026.0329.1
sudoLMBackbone=Mistral 7B2026.0323.85
LLaMA2Model Size=7B, Method=Few-shot, Prompt=General2024.0315.2
sudoLMBackbone=Llama3.1 8B2026.0314.76