List your company

Question Answering on Winogrande (WG)

74.1AccuracyFP16
56.05660.740565.42570.1095Nov 21, 2024Feb 14, 2025May 11, 2025Aug 5, 2025Oct 30, 2025Jan 24, 2026Apr 20, 2026

Evaluation results

MethodLinks
FP16Model=LLaMA3.1-8B-Instruct, Precision=MXFP42026.0474.1
LLaMA-3 8BPruning Ratio=0%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0272.77
DuQuant++Model=LLaMA3-8B-Instruct, Precision=MXFP42026.0472.1
DuQuant++*Model=LLaMA3.1-8B-Instruct, Precision=MXFP42026.0471.7
DuQuant++*Model=LLaMA3-8B-Instruct, Precision=MXFP42026.0471.6
FP16Model=LLaMA3-8B-Instruct, Precision=MXFP42026.0471.4
FlatQuantModel=LLaMA3-8B-Instruct, Precision=MXFP42026.0471.1
FlatQuantModel=LLaMA3.1-8B-Instruct, Precision=MXFP42026.0471.1
CoMePruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0270.96
QuaRot*Model=LLaMA3.1-8B-Instruct, Precision=MXFP42026.0470.7
MR-GPTQModel=LLaMA3.1-8B-Instruct, Precision=MXFP42026.0470.6
DuQuant++Model=LLaMA3.1-8B-Instruct, Precision=MXFP42026.0470.6
LINEARPATCHPruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0270.17
HPTQAvg. Bitwidth=4.125, Backbone=Llama-2-7B, Zero-shot=true2025.0769.61
FP16 BaselineAvg. Bitwidth=16.000, Backbone=Llama-2-7B, Zero-shot=true2025.0769.46
LLaMA-2 7BPruning Ratio=0%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0269.14
MR-GPTQModel=LLaMA3-8B-Instruct, Precision=MXFP42026.0469
SSQR-1%Avg. Bitwidth=4.445, Backbone=Llama-2-7B, Zero-shot=true2025.0768.82
GPTQAvg. Bitwidth=4.125, Backbone=Llama-2-7B, Zero-shot=true2025.0768.82
QuaRot*Model=LLaMA3-8B-Instruct, Precision=MXFP42026.0468.8
QuaRotModel=LLaMA3.1-8B-Instruct, Precision=MXFP42026.0468.4
HPTQAvg. Bitwidth=3.125, Backbone=Llama-2-7B, Zero-shot=true2025.0767.72
QuIP#Avg. Bitwidth=4.000, Backbone=Llama-2-7B, Zero-shot=true2025.0767.6
LINEARPATCHPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0267.4
AQLMAvg. Bitwidth=5.020, Backbone=Llama-2-7B, Zero-shot=true2025.0767.4
AQLMAvg. Bitwidth=4.040, Backbone=Llama-2-7B, Zero-shot=true2025.0767.32
QuaRotModel=LLaMA3-8B-Instruct, Precision=MXFP42026.0467.3
CoMePruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0267.25
QTIPAvg. Bitwidth=4.000, Backbone=Llama-2-7B, Zero-shot=true2025.0767.1
AQLMAvg. Bitwidth=3.040, Backbone=Llama-2-7B, Zero-shot=true2025.0766.93
AmoebaLLMPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0266.9
QTIPAvg. Bitwidth=3.000, Backbone=Llama-2-7B, Zero-shot=true2025.0766.9
MoPPruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0266.8
LINEARPATCHPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0266.69
QuIP#Avg. Bitwidth=3.000, Backbone=Llama-2-7B, Zero-shot=true2025.0766.5
PruneNetPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0266.22
HPTQAvg. Bitwidth=2.125, Backbone=Llama-2-7B, Zero-shot=true2025.0765.82
AQLMAvg. Bitwidth=2.290, Backbone=Llama-2-7B, Zero-shot=true2025.0765.67
AQLMAvg. Bitwidth=2.020, Backbone=Llama-2-7B, Zero-shot=true2025.0765.67
SSQR-1%Avg. Bitwidth=3.445, Backbone=Llama-2-7B, Zero-shot=true2025.0765.43
MoPPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0265.19
Pretrained ModelModel Size=7B, Training Strategy=Standard Pretraining2026.0365.19
SliceGPTPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0265.11
GPTQAvg. Bitwidth=3.125, Backbone=Llama-2-7B, Zero-shot=true2025.0764.96
QuIP#Avg. Bitwidth=2.000, Backbone=Llama-2-7B, Zero-shot=true2025.0764.9
SlimLLMPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0264.88
QTIPAvg. Bitwidth=2.000, Backbone=Llama-2-7B, Zero-shot=true2025.0764.7
ModHiFi-PPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0264.64
DISP-LLMPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0263.93
Pretrained ModelModel Size=2.7B, Training Strategy=Standard Pretraining2026.0363.85
MoPPruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0263.75
CoMePruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0263.38
Split Model TrainingModel Size=2.7B, Training Strategy=Split Training2026.0363.22
PruneNetPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0262.9
MoPPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0262.38
Split Model TrainingModel Size=1.3B, Training Strategy=Split Training2026.0362.19
ProDiaLModel Size=Mamba-1.4B, Parameters=0.05B, Projection Type=Both-Proj2024.1161.96
Pretrained ModelModel Size=1.3B, Training Strategy=Standard Pretraining2026.0361.88
StrongModel Size=Mamba-1.4B, Parameters=0.06B, Projection Type=Base2024.1161.8
ProDiaLModel Size=Mamba-1.4B, Parameters=0.02B, Projection Type=Out-Proj2024.1161.8
FTModel Size=Mamba-1.4B, Parameters=1.21B, Projection Type=Both-Proj2024.1161.72
LoRAModel Size=Mamba-1.4B, Parameters=0.05B, Projection Type=Both-Proj2024.1161.72
DoRAModel Size=Mamba-1.4B, Parameters=0.05B, Projection Type=Both-Proj2024.1161.72
FTModel Size=Mamba-1.4B, Parameters=0.40B, Projection Type=Out-Proj2024.1161.72
FTModel Size=Mamba-1.4B, Parameters=0.81B, Projection Type=In-Proj2024.1161.56
LoRAModel Size=Mamba-1.4B, Parameters=0.03B, Projection Type=In-Proj2024.1161.56
ProDiaLModel Size=Mamba-1.4B, Parameters=0.03B, Projection Type=In-Proj2024.1161.56
LoRAModel Size=Mamba-1.4B, Parameters=0.02B, Projection Type=Out-Proj2024.1161.56
AMPPruning Ratio=20%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0261.56
DoRAModel Size=Mamba-1.4B, Parameters=0.03B, Projection Type=In-Proj2024.1161.48
DoRAModel Size=Mamba-1.4B, Parameters=0.02B, Projection Type=Out-Proj2024.1161.48
MoPPruning Ratio=40%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0261.38
SliceGPTPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0261.33
BitFitModel Size=Mamba-1.4B, Parameters=0.39M, Projection Type=Base2024.1161.25
AMPPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0261.25
Full-FTModel Size=Mamba-1.4B, Parameters=1.40B, Projection Type=Base2024.1161.17
Shortened LLaMAPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0261.09
Yang et al.Pruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0260.77
AMPPruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0260.77
CoMePruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0260.62
Yang et al.Pruning Ratio=30%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0260.3
AMPPruning Ratio=20%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0259.91
ModHiFi-PPruning Ratio=30%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0259.35
MoPPruning Ratio=40%, Base Model=LLaMA-2 7B, Evaluation Protocol=Zero-shot2026.0259.35
ProDiaLModel Size=Mamba-370M, Parameters=2.02M, Projection Type=Out-Proj2024.1157.3
FTModel Size=Mamba-370M, Parameters=302M, Projection Type=Both-Proj2024.1157.22
DoRAModel Size=Mamba-370M, Parameters=4.13M, Projection Type=In-Proj2024.1157.22
ProDiaLModel Size=Mamba-370M, Parameters=3.74M, Projection Type=In-Proj2024.1157.22
FTModel Size=Mamba-370M, Parameters=101M, Projection Type=Out-Proj2024.1157.22
DoRAModel Size=Mamba-370M, Parameters=2.41M, Projection Type=Out-Proj2024.1157.22
StrongModel Size=Mamba-370M, Parameters=8.19M, Projection Type=Base2024.1157.14
ProDiaLModel Size=Mamba-370M, Parameters=5.75M, Projection Type=Both-Proj2024.1157.06
FTModel Size=Mamba-370M, Parameters=201M, Projection Type=In-Proj2024.1157.06
Full-FTModel Size=Mamba-370M, Parameters=370M, Projection Type=Base2024.1156.99
BitFitModel Size=Mamba-370M, Parameters=0.20M, Projection Type=Base2024.1156.99
DoRAModel Size=Mamba-370M, Parameters=6.54M, Projection Type=Both-Proj2024.1156.99
LoRAModel Size=Mamba-370M, Parameters=2.36M, Projection Type=Out-Proj2024.1156.91
AMPPruning Ratio=40%, Base Model=LLaMA-3 8B, Evaluation Protocol=Zero-shot2026.0256.91
LoRAModel Size=Mamba-370M, Parameters=6.29M, Projection Type=Both-Proj2024.1156.75
LoRAModel Size=Mamba-370M, Parameters=3.93M, Projection Type=In-Proj2024.1156.75