List your company

Zero-shot Reasoning on PIQA

82.54PIQA Zero-shot AccuracyFP16
55.978462.874269.7776.6658Jul 25, 2023Jan 14, 2024Jul 6, 2024Dec 27, 2024Jun 19, 2025Dec 10, 2025Jun 2, 2026

Evaluation results

MethodLinks
FP16Model=Mistral, Precision=FP16, Sequence Length=20482025.0782.54
FP16Backbone=Llama-3-70B, Bits=FP162026.0682.37
LiftQuantBackbone=Llama-2-70B, Bits=2.412026.0682.32
QTIPBackbone=Llama-2-70B, Bits=2.002026.0682.16
FP16Backbone=Llama-2-70B, Bits=FP162026.0682.15
OASISModel=Mistral, Precision=W4A4, Sequence Length=20482025.0782.1
OASIS-SModel=Mistral, Precision=W4A4, Sequence Length=20482025.0781.77
LiftQuantBackbone=Llama-3-70B, Bits=2.012026.0681.72
QuIP#Backbone=Llama-2-70B, Bits=2.002026.0681.39
LiftQuantBackbone=Llama-2-70B, Bits=2.012026.0681.28
OPTQWBits=162023.0780.9
QuIPWBits=162023.0780.9
AtomModel=Mistral, Precision=W4A4, Sequence Length=2048, Group Size=1282025.0780.71
FP16Model=LLaMA-3-8B, Precision=FP16, Sequence Length=20482025.0780.63
OPTQWBits=42023.0780.52
QuIPWBits=42023.0780.47
VPTQBackbone=Llama-3-70B, Bits=2.022026.0680.36
QuIPWBits=32023.0780.25
QuaRotModel=Mistral, Precision=W4A4, Sequence Length=20482025.0780.19
EQAT-g64Backbone=Llama-2-70B, Bits=2.262026.0680.14
OASISModel=Mistral, Precision=W4A3, Sequence Length=20482025.0779.87
FP16Backbone=Llama-3-8B, Bits=FP162026.0679.6
FP16Backbone=Llama-2-13B, Bits=FP162026.0679.05
BaselineRatio=1.0, Memory Footprint (GB)=12.6GB, Zero-shot=true2026.0479
FP16Model=LLaMA-2-7B, Precision=FP16, Sequence Length=20482025.0778.67
OASISModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=20482025.0778.67
Llama2-7BBackbone=Llama2-7B2026.0278.6
OPTQWBits=32023.0778.56
FP16Backbone=Llama-2-7B, Bits=FP162026.0678.07
DensePrune Rate=0%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0678.02
OASISModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=20482025.0777.97
QuIP#Backbone=Llama-2-13B, Bits=2.002026.0677.97
LiftQuantBackbone=Llama-2-13B, Bits=2.022026.0677.91
QTIPBackbone=Llama-2-13B, Bits=2.002026.0677.69
OASIS-SModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=20482025.0777.62
LiftQuantBackbone=Llama-2-13B, Bits=2.422026.0677.48
EQAT-g64Backbone=Llama-3-70B, Bits=2.252026.0677.37
OASIS-SModel=Mistral, Precision=W4A3, Sequence Length=20482025.0777.35
OASIS-SModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=20482025.0777.31
OASISModel=LLaMA-3-8B, Precision=W4A3, Sequence Length=20482025.0777.09
EQAT-g64Backbone=Llama-2-13B, Bits=2.252026.0677.04
LiftQuantBackbone=Llama-2-7B, Bits=2.422026.0676.99
LiftQuantBackbone=Llama-2-7B, Bits=2.022026.0676.55
LiftQuantBackbone=Llama-3-8B, Bits=2.022026.0676.55
COMPOTBackbone=Llama2-7B, Training-free=true, CR=0.22026.0276.4
QuaRotModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=20482025.0776.39
QTIPBackbone=Llama-2-7B, Bits=2.002026.0676.28
EQAT-g64Backbone=Llama-3-8B, Bits=2.252026.0676.03
OASISModel=LLaMA-2-7B, Precision=W4A3, Sequence Length=20482025.0775.84
OASIS-SModel=LLaMA-3-8B, Precision=W4A3, Sequence Length=20482025.0775.82
PG pruningPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0675.41
EQAT-g64Backbone=Llama-2-7B, Bits=2.252026.0675.3
AtomModel=LLaMA-2-7B, Precision=W4A4, Sequence Length=2048, Group Size=1282025.0775.14
OASIS-SModel=LLaMA-2-7B, Precision=W4A3, Sequence Length=20482025.0775.14
VPTQBackbone=Llama-3-8B, Bits=2.072026.0675.12
QuIPWBits=22023.0775.08
QuIP#Backbone=Llama-2-7B, Bits=2.002026.0674.16
BonsaiPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0673.94
AtomModel=Mistral, Precision=W4A3, Sequence Length=2048, Group Size=1282025.0773.69
Nirvana (Ours)Setting=Zero-shot, Model Size=1.3B parameters2025.1073.67
Nirvana-noTriggerSetting=Zero-shot, Model Size=1.3B parameters2025.1073.12
Swift-SVDRatio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0473
Swift-SVD*Ratio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0473
AtomModel=LLaMA-3-8B, Precision=W4A3, Sequence Length=2048, Group Size=1282025.0772.86
Gated DeltaNet-H1Setting=Zero-shot, Model Size=1.3B parameters2025.1072.57
SliceGPTPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0672.31
Gated DeltaNetSetting=Zero-shot, Model Size=1.3B parameters2025.1072.25
Gated DeltaNet-H2Setting=Zero-shot, Model Size=1.3B parameters2025.1072.19
Dobi-SVD(w)Ratio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0472
Mamba2Setting=Zero-shot, Model Size=1.3B parameters2025.1071.87
LLM-PrunerPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0671.81
Wanda-spPrune Rate=30%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0671.6
MambaSetting=Zero-shot, Model Size=1.3B parameters2025.1071.32
PG pruningPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0671.11
AtomModel=LLaMA-2-7B, Precision=W4A3, Sequence Length=2048, Group Size=1282025.0771.01
SVD-LLM(W)Ratio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0471
SambaSetting=Zero-shot, Model Size=1.3B parameters2025.1070.94
DeltaNetSetting=Zero-shot, Model Size=1.3B parameters2025.1070.72
HGRN2Setting=Zero-shot, Model Size=1.3B parameters2025.1070.45
RetNetSetting=Zero-shot, Model Size=1.3B parameters2025.1070.07
Transformer++Setting=Zero-shot, Model Size=1.3B parameters2025.1070.02
AtomModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=2048, Group Size=1282025.0769.45
BonsaiPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0668.44
QuaRotModel=LLaMA-3-8B, Precision=W4A4, Sequence Length=20482025.0768.28
ASVDRatio=0.8, Memory Footprint (GB)=10.1 GB, Zero-shot=true2026.0468
LLM-PrunerPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0667.52
COMPOTBackbone=Llama2-7B, Training-free=true, CR=0.42026.0267
Swift-SVD*Ratio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0467
Swift-SVDRatio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0466
SliceGPTPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0665.4
Dobi-SVDBackbone=Llama2-7B, Training-free=false, CR=0.22026.0265.2
Wanda-spPrune Rate=40%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0664.63
Dobi-SVD(w)Ratio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0464
SVD-LLM(W)Ratio=0.6, Memory Footprint (GB)=7.7 GB, Zero-shot=true2026.0463
PG pruningPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0661.8
LLM-PrunerPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0659.52
BonsaiPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0659.52
SliceGPTPrune Rate=50%, Backbone=LLaMA-2-7B, Evaluation Protocol=Zero-shot2024.0659.47
Dobi-SVDBackbone=Llama2-7B, Training-free=false, CR=0.42026.0257.3
Swift-SVD*Ratio=0.4, Memory Footprint (GB)=5.3 GB, Zero-shot=true2026.0457