List your company
ResearchFormerly WizWand

Causal Reasoning on COPA

95AccuracyFew-shot*
58.668.0577.586.95Mar 30, 2023Oct 6, 2023Apr 14, 2024Oct 22, 2024May 1, 2025Nov 8, 2025May 18, 2026
Updated 5mo ago

Evaluation results

MethodLinks
Few-shot*Backbone=Qwen2.5-7B, Few-shot settings=3-shot OOD2025.0995——
ICRBackbone=Qwen2.5-7B2025.0995——
M²IVBackbone=Qwen2.5-7B2025.0993.5——
LIVEBackbone=Qwen2.5-7B2025.0993.2——
I2CLBackbone=Qwen2.5-7B2025.0992.6——
Zero-shotBackbone=Qwen2.5-7B2025.0992——
LATModel=microsoft/deberta-xxlarge-v2-mnli2023.1090——
GPT-NeoXEvaluation Protocol=1-shot2023.0388——
GPT-3Evaluation Protocol=1-shot2023.0387——
DenseBase Model=LLaMA-2-7B, Sparsity=Dense2025.0687——
MaskProBase Model=DeepSeek-7B, Sparsity=2:42025.0687——
BloombergGPTEvaluation Protocol=1-shot2023.0386——
OPT-66BEvaluation Protocol=1-shot2023.0386——
SparsegptBase Model=DeepSeek-7B, Sparsity=2:42025.0686——
BLOOM-176BEvaluation Protocol=1-shot2023.0384——
DenseBase Model=DeepSeek-7B, Sparsity=Dense2025.0684——
Pruner-ZBase Model=DeepSeek-7B, Sparsity=2:42025.0684——
UniBiasModel=Llama-2 13b, ICL Setting=1-shot2024.0583.2——
Few-shot*Backbone=Llama2-7B, Few-shot settings=3-shot OOD2025.0982——
SparsegptBase Model=LLaMA-2-7B, Sparsity=2:42025.0681——
FIXED_E=128Experts=1282026.0581——
EMO (Stage 5)E=64→1282026.0580——
Our Trained ModelModel Size=52B, Data type=FP16, Zero-shot=true2023.0579.4——
Our Trained ModelModel Size=52B, Data type=W8A8, Zero-shot=true2023.0579.4——
Pruner-ZBase Model=LLaMA-2-7B, Sparsity=2:42025.0679——
MaskProBase Model=LLaMA-2-7B, Sparsity=2:42025.0679——
Our Trained ModelModel Size=52B, Data type=W4, Zero-shot=true2023.0578.2——
PCModel=Llama-2 13b, ICL Setting=1-shot2024.0576.8——
Our Trained ModelModel Size=13B, Data type=FP16, Zero-shot=true2023.0576——
Our Trained ModelModel Size=13B, Data type=W8A8, Zero-shot=true2023.0576——
FIXED_E=32Experts=322026.0576——
Our Trained ModelModel Size=6B, Data type=FP16, Zero-shot=true2023.0575.4——
CCModel=Llama-2 13b, ICL Setting=1-shot2024.0575.2——
Our Trained ModelModel Size=6B, Data type=W8A8, Zero-shot=true2023.0574.6——
Our Trained ModelModel Size=6B, Data type=W4, Zero-shot=true2023.0574——
FIXED_E=16Experts=162026.0574——
EMO (Stage 3)E=16→322026.0574——
Our Trained ModelModel Size=13B, Data type=W4, Zero-shot=true2023.0573.6——
EMO (Stage 2)E=8→162026.0573——
SYNPROModel Scale=1.1B, Organic Tokens=2.2B2026.0573——
ReProModel Scale=1.1B, Organic Tokens=2.2B2026.0572——
DCModel=Llama-2 13b, ICL Setting=1-shot2024.0571——
SYNPROModel Scale=400M, Organic Tokens=0.8B2026.0570——
UniBiasModel=Llama-2 7b, ICL Setting=1-shot2024.0569——
ReProModel Scale=400M, Organic Tokens=0.8B2026.0569——
EMO (Stage 1)E=82026.0568——
EMO (Stage 4)E=32→642026.0568——
ICRBackbone=Llama2-7B2025.0968——
CCModel=Llama-2 7b, ICL Setting=1-shot2024.0567.8——
PCModel=Llama-2 7b, ICL Setting=1-shot2024.0567.8——
ICLModel=Llama-2 7b, ICL Setting=1-shot2024.0567.6——
ICLModel=Llama-2 13b, ICL Setting=1-shot2024.0567.5——
Our Trained ModelModel Size=410M, Data type=FP16, Zero-shot=true2023.0565.4——
LIVEBackbone=Llama2-7B2025.0964.8——
M²IVBackbone=Llama2-7B2025.0964.8——
Our Trained ModelModel Size=410M, Data type=W8A8, Zero-shot=true2023.0564——
I2CLBackbone=Llama2-7B2025.0963.6——
RepeatModel Scale=1.1B, Organic Tokens=2.2B2026.0563——
Zero-shotBackbone=Llama2-7B2025.0963——
Our Trained ModelModel Size=410M, Data type=W4, Zero-shot=true2023.0562.8——
CCSModel=microsoft/deberta-xxlarge-v2-mnli2023.1061——
DCModel=Llama-2 7b, ICL Setting=1-shot2024.0560.4——
RepeatModel Scale=400M, Organic Tokens=0.8B2026.0560——
OPT 1.3BScale=1.3B2022.12—7775
OPT 175BScale=175B2022.12—8791.4
OPT 30BScale=30B2022.12—8488.6
OPT-IML 1.3BScale=1.3B2022.12—7778
OPT-IML 175BScale=175B2022.12—8889
OPT-IML 30BScale=30B2022.12—8589