List your company
ResearchFormerly WizWand

Reading Comprehension on MultiRC (Accuracy)

72.9MultiRC AccuracyGPT-3
16.63631.24345.8560.457Dec 2, 2022Jul 5, 2023Feb 6, 2024Sep 8, 2024Apr 12, 2025Nov 13, 2025Jun 17, 2026
Updated 4mo ago

Evaluation results

MethodLinks
GPT-3Shots=12023.0372.9
MUPPET2022.1264.15
ColD-Fusion2022.1263.01
Multitask2022.1262.34
BloombergGPTShots=12023.0362.29
Finetune2022.1260.56
DenseBase Model=DeepSeek-7B, Sparsity=Dense2025.0657.22
SparsegptBase Model=LLaMA-2-7B, Sparsity=2:42025.0657.2
SparsegptBase Model=DeepSeek-7B, Sparsity=2:42025.0657.2
Pruner-ZBase Model=DeepSeek-7B, Sparsity=2:42025.0657.2
MaskProBase Model=DeepSeek-7B, Sparsity=2:42025.0657.2
DenseBase Model=LLaMA-2-7B, Sparsity=Dense2025.0656.97
MaskProBase Model=LLaMA-2-7B, Sparsity=2:42025.0656.93
RegMixBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=3.072 × 10^182026.0456.7
UniformBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=02026.0456.68
ADAPT-BM25Backbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=≪ 1.0 × 10^142026.0456.53
Pruner-ZBase Model=LLaMA-2-7B, Sparsity=2:42025.0656.37
ADAPTBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=≪ 1.1 × 10^152026.0455.67
Human2026.0654.15
DoReMi2026.0653.31
RegMix2026.0652.86
RegMix-DProxy models=512, Mode=Offline2026.0652.13
RegMix-DProxy models=512, Mode=Online2026.0651.71
RegMix-DProxy models=128, Mode=Online2026.0651.57
LinUpperBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=02026.0450.74
RegMix-DProxy models=128, Mode=Offline2026.0650.25
DoReMiBackbone=TinyLlama-120M, Training Budget=50B tokens, FLOPs overhead=4.92 × 10^192026.0448.99
SSAArchitecture=Nemotron-style decoder, Parameters=114M, Training steps=22K, Training dataset=FineWeb, Zero-shot=true2025.0843.5
SoftmaxArchitecture=Nemotron-style decoder, Parameters=114M, Training steps=22K, Training dataset=FineWeb, Zero-shot=true2025.0842.8
BLOOM176BShots=12023.0326.65
GPT-NeoXShots=12023.0322.86
OPT66BShots=12023.0318.8