List your company

Multiple Choice Question Answering on OpenBookQA

44.4AccuracyOriginal
14.86422.53230.237.868Jan 13, 2026Feb 9, 2026Mar 8, 2026Apr 5, 2026May 2, 2026May 29, 2026Jun 26, 2026

Evaluation results

MethodLinks
OriginalBackbone=Mistral-7B, Pruning Ratio=0%2026.0544.4
AccBackbone=Mistral-7B, Pruning Ratio=25%, Relevance Estimation Source=Task-specific2026.0542.4
Out. Cosine-SimBackbone=Mistral-7B, Pruning Ratio=25%, Relevance Estimation Source=Task-specific2026.0537.8
Out. Norm-SimBackbone=Mistral-7B, Pruning Ratio=25%, Relevance Estimation Source=Task-specific2026.0537.8
Cosine SimilarityBackbone=Mistral-7B, Pruning Ratio=25%, Relevance Estimation Source=Task-specific2026.0537.4
ExoFormerAnchor Source=External Anchor, Mixing Granularity=Elementwise, Normalization Strategy=Full2026.0136.4
PerplexityBackbone=Mistral-7B, Pruning Ratio=25%, Relevance Estimation Source=Task-specific2026.0536.4
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Elementwise, Normalization Strategy=No Norm, Mixing Path Components={V,G}2026.0135
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Scalar, Normalization Strategy=No Norm, Mixing Path Components={V,G}2026.0134.8
ResFormer (Value Residual)Anchor Source=Baselines2026.0134.4
ExoFormerAnchor Source=External Anchor, Mixing Granularity=Elementwise, Normalization Strategy=No Norm2026.0134.4
Dynamic ExoFormerAnchor Source=External Anchor, Mixing Granularity=Elementwise, Normalization Strategy=Full2026.0134.4
Out. Divergence-SimBackbone=Mistral-7B, Pruning Ratio=25%, Relevance Estimation Source=Task-specific2026.0534.4
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Headwise, Normalization Strategy=Only Q/K Norms2026.0134.2
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Scalar, Normalization Strategy=Only Q/K Norms2026.0134.2
Gated AttentionAnchor Source=Baselines2026.0134
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Headwise, Normalization Strategy=No Norm, Mixing Path Components={V,G}2026.0134
Base TransformerAnchor Source=Baselines2026.0133.6
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Elementwise, Normalization Strategy=Full2026.0133.6
ExoFormerAnchor Source=External Anchor, Mixing Granularity=Scalar, Normalization Strategy=Full2026.0133.6
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Headwise, Normalization Strategy=Full2026.0133.4
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Scalar, Normalization Strategy=Full2026.0133.2
NuResFormerAnchor Source=Internal Anchor, Mixing Granularity=Elementwise, Normalization Strategy=Only Q/K Norms2026.0133
Naïve CombinationAnchor Source=Baselines2026.0132.6
ExoFormerAnchor Source=External Anchor, Mixing Granularity=Headwise, Normalization Strategy=Full2026.0132.4
R2LMShots=0-sh, Candidate Type=long-target2026.0627.4
Causal dLLMShots=0-sh, Candidate Type=long-target2026.0627
R2LM-PI (plug-in)Shots=0-sh, Candidate Type=long-target2026.0625.6
TaylorBackbone=Mistral-7B, Pruning Ratio=25%, Relevance Estimation Source=Task-specific2026.0524.4
Bidirectional dLLMShots=0-sh, Candidate Type=long-target2026.0616