List your company

Zero-shot Reasoning on WinoGrande

80.51AccuracyFP16
47.510856.077964.64573.2121Jun 15, 2024Oct 12, 2024Feb 9, 2025Jun 8, 2025Oct 6, 2025Feb 2, 2026Jun 2, 2026

Evaluation results

MethodLinks
FP16Backbone=Llama-3-70B, Bits=FP162026.0680.51
LiftQuantBackbone=Llama-3-70B, Bits=2.012026.0678.53
FP16Backbone=Llama-2-70B, Bits=FP162026.0677.98
VPTQBackbone=Llama-3-70B, Bits=2.022026.0677.9
LiftQuantBackbone=Llama-2-70B, Bits=2.412026.0677.19
QTIPBackbone=Llama-2-70B, Bits=2.002026.0676.19
QuIP#Backbone=Llama-2-70B, Bits=2.002026.0675.77
LiftQuantBackbone=Llama-2-70B, Bits=2.012026.0675.69
FP16Model=Mistral, Precision=FP16, Sequence Length=20482025.0775.22
EQAT-g64Backbone=Llama-2-70B, Bits=2.262026.0674.59
EQAT-g64Backbone=Llama-3-70B, Bits=2.252026.0674.03
FP16Model=LLaMA-3-8B, Precision=FP16, Sequence Length=20482025.0773.7
OASISModel=Mistral, Precision=W4A4, Sequence Length=20482025.0773.4
OASIS-SModel=Mistral, Precision=W4A4, Sequence Length=20482025.0772.97
FP16Backbone=Llama-3-8B, Bits=FP162026.0672.61
QuaRotModel=Mistral, Precision=W4A4, Sequence Length=20482025.0772.34
FP16Backbone=Llama-2-13B, Bits=FP162026.0672.22
AtomModel=Mistral, Precision=W4A4, Sequence Length=2048, Group Size=1282025.0772.03
OASISModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=20482025.0771.27
LiftQuantBackbone=Llama-2-13B, Bits=2.422026.0671.27
OASIS-SModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=20482025.0770.56
OASISModel=Mistral, Precision=W4A3, Sequence Length=20482025.0770.56
QTIPBackbone=Llama-2-13B, Bits=2.002026.0670.4
LiftQuantBackbone=Llama-2-13B, Bits=2.022026.0670.01
BaselineRatio=1.0, Memory Footprint (GB)=12.6GB, Zero-shot=true2026.0470
FP16Backbone=Llama-2-7B, Bits=FP162026.0669.22
QuIP#Backbone=Llama-2-13B, Bits=2.002026.0669.06
Llama2-7BBackbone=Llama2-7B2026.0269
FP16Model=LLaMA-2-7B, Precision=FP16, Sequence Length=20482025.0768.98
DensePrune Rate=0%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0668.43
EQAT-g64Backbone=Llama-2-13B, Bits=2.252026.0668.36
LiftQuantBackbone=Llama-2-7B, Bits=2.022026.0668.27
LiftQuantBackbone=Llama-3-8B, Bits=2.022026.0668.03
Swift-SVDRatio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0468
Swift-SVD*Ratio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0468
LiftQuantBackbone=Llama-2-7B, Bits=2.422026.0667.72
EQAT-g64Backbone=Llama-3-8B, Bits=2.252026.0667.72
COMPOTBackbone=Llama2-7B, Training-free=true, CR=0.22026.0267.6
QTIPBackbone=Llama-2-7B, Bits=2.002026.0667.25
OASIS-SModel=Mistral, Precision=W4A3, Sequence Length=20482025.0766.93
OASISModel=LLaMA-3-8B, Precision=W4A3, Sequence Length=20482025.0766.22
EQAT-g64Backbone=Llama-2-7B, Bits=2.252026.0665.98
VPTQBackbone=Llama-3-8B, Bits=2.072026.0665.92
QuIP#Backbone=Llama-2-7B, Bits=2.002026.0665.67
OASISModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=20482025.0765.51
OASIS-SModel=LLaMA-3-8B, Precision=W4A3, Sequence Length=20482025.0765.03
OASIS-SModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=20482025.0764.8
OASISModel=LLaMA-2-7B, Precision=W4A3, Sequence Length=20482025.0764.17
ASVDRatio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0464
OASIS-SModel=LLaMA-2-7B, Precision=W4A3, Sequence Length=20482025.0763.93
AtomModel=Mistral, Precision=W4A3, Sequence Length=2048, Group Size=1282025.0763.73
COMPOTBackbone=Llama2-7B, Training-free=true, CR=0.42026.0263.5
SliceGPTPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0663.22
QuaRotModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=20482025.0763.06
QuaRotModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=20482025.0763.06
AtomModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=2048, Group Size=1282025.0762.75
Dobi-SVD(w)Ratio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0462
Swift-SVDRatio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0462
Swift-SVD*Ratio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0462
PG pruningPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0661.6
AtomModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=2048, Group Size=1282025.0761.13
SVD-LLM(W)Ratio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0461
BonsaiPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0660.93
AtomModel=LLaMA-3-8B, Precision=W4A3, Sequence Length=2048, Group Size=1282025.0760.87
SliceGPTPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0660.38
Wanda-spPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0660.3
AtomModel=LLaMA-2-7B, Precision=W4A3, Sequence Length=2048, Group Size=1282025.0759.5
Nirvana (Ours)Setting=Zero-shot, Model Size=1.3B parameters2025.1059.48
Nirvana-noTriggerSetting=Zero-shot, Model Size=1.3B parameters2025.1059.27
Dobi-SVDBackbone=Llama2-7B, Training-free=false, CR=0.22026.0258.5
Gated DeltaNet-H1Setting=Zero-shot, Model Size=1.3B parameters2025.1058.4
Dobi-SVD(w)Ratio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0458
Gated DeltaNet-H2Setting=Zero-shot, Model Size=1.3B parameters2025.1057.77
Gated DeltaNetSetting=Zero-shot, Model Size=1.3B parameters2025.1057.45
SliceGPTPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0656.27
PG pruningPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0655.72
SambaSetting=Zero-shot, Model Size=1.3B parameters2025.1055.56
BonsaiPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0655.41
Mamba2Setting=Zero-shot, Model Size=1.3B parameters2025.1055.24
SVD-LLM(W)Ratio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0455
Dobi-SVDBackbone=Llama2-7B, Training-free=false, CR=0.42026.0254.4
RetNetSetting=Zero-shot, Model Size=1.3B parameters2025.1054.14
LLM-PrunerPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0654.06
Transformer++Setting=Zero-shot, Model Size=1.3B parameters2025.1053.51
COMPOTBackbone=Llama2-7B, Training-free=true, CR=0.62026.0253.4
DeltaNetSetting=Zero-shot, Model Size=1.3B parameters2025.1053.35
BonsaiPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0653.12
Swift-SVD*Ratio=0.4, Memory Footprint (GB)=5.3 GB, Zero-shot=true2026.0453
MambaSetting=Zero-shot, Model Size=1.3B parameters2025.1052.95
Wanda-spPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0652.8
HGRN2Setting=Zero-shot, Model Size=1.3B parameters2025.1052.8
PG pruningPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0652.64
Wanda-spPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0652.17
Swift-SVDRatio=0.4, Memory Footprint (GB)=5.3 GB, Zero-shot=true2026.0452
LLM-PrunerPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0651.7
Dobi-SVDBackbone=Llama2-7B, Training-free=false, CR=0.62026.0250.5
QuaRotModel=LLaMA-3-8B, Precision=W4A3, Sequence Length=20482025.0750.2
LLM-PrunerPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0650.11
QuaRotModel=LLaMA-2-7B, Precision=W4A3, Sequence Length=20482025.0749.49
QuaRotModel=Mistral, Precision=W4A3, Sequence Length=20482025.0748.78