List your company

Zero-shot Evaluation on PIQA, WinoGrande, HellaSwag, ARC (Easy and Challenge), and LAMBADA (test)

77.07Average AccuracyFP16
33.826845.053456.2867.5066Mar 30, 2024Aug 15, 2024Dec 31, 2024May 18, 2025Oct 3, 2025Feb 18, 2026Jul 7, 2026

Evaluation results

MethodLinks
FP16Model=LLAMA2-70B, Quantization=FP16, Evaluation Protocol=Zero-shot2024.0377.0782.777.9883.8480.9857.3479.58
QuaRotModel=LLAMA2-70B, Quantization=4-bit (A4W4KV4), Evaluation Protocol=Zero-shot2024.0375.9882.4376.2481.8280.4356.2378.73
16bitsAvg. bits=16, Group Size=-, Backbone=Llama3-70B2025.1275.28——————
QuaRot-GPTQModel=Phi-3-mini, Precision=INT82024.0373.1880.2574.1978.5480.3557.0868.64
BaselineModel=Phi-3-mini, Precision=FP162024.0373.1180.4773.7278.4580.1357.5168.37
QuaRot-RTNModel=Phi-3-mini, Precision=INT82024.0373.0279.7174.1178.6380.4756.6668.56
16bitsAvg. bits=16, Group Size=-, Backbone=Llama2-70B2025.1272.71——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=128, Backbone=Llama3-70B2025.1272.68——————
QuaRot-GPTQModel=Phi-3-mini, Precision=INT62024.0372.6779.5472.6978.579.4256.7468.85
FP16Model=LLAMA2-13B, Quantization=FP16, Evaluation Protocol=Zero-shot2024.0372.5980.4772.2279.3977.4849.2376.75
SignRoundV2Avg. bits=2.5, Group Size=128, Backbone=Llama3-70B2025.1272.37——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=64, Backbone=Llama3-70B2025.1272.32——————
QuaRot-RTNModel=Phi-3-mini, Precision=INT62024.0371.9879.5473.0177.4679.2155.1267.53
SignRoundV2Avg. bits=2.5, Group Size=64, Backbone=Llama3-70B2025.1271.97——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=64, Backbone=Llama3-70B2025.1270.93——————
QuIP#Avg. bits=2, Group Size=-, Backbone=Llama2-70B2025.1270.91——————
AQLMAvg. bits=2, Group Size=1x16, Backbone=Llama2-70B2025.1270.84——————
SignRoundV2Avg. bits=2.5, Group Size=128, Backbone=Llama2-70B2025.1270.6——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=64, Backbone=Llama2-70B2025.1270.6——————
SignRoundV2Avg. bits=2.5, Group Size=64, Backbone=Llama2-70B2025.1270.25——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=128, Backbone=Llama3-70B2025.1270.16——————
AQLMAvg. bits=2, Group Size=1x16, Backbone=Llama3-70B2025.1270.1——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=128, Backbone=Llama2-70B2025.1270.04——————
AQLMAvg. bits=2, Group Size=2x8, Backbone=Llama2-70B2025.1269.85——————
FP16Model=LLAMA2-7B, Quantization=FP16, Evaluation Protocol=Zero-shot2024.0369.8279.1169.0675.9974.5846.2573.9
QuaRotModel=LLAMA2-13B, Quantization=4-bit (A4W4KV4), Evaluation Protocol=Zero-shot2024.0369.7978.8970.2476.3772.9846.5973.67
EfficientQATAvg. bits=2, Group Size=64, Backbone=Llama2-70B2025.1269.48——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=64, Backbone=Llama2-70B2025.1269.3——————
SignRoundV2Avg. bits=2, Group Size=128, Backbone=Llama3-70B2025.1269.02——————
EfficientQATAvg. bits=2, Group Size=128, Backbone=Llama2-70B2025.1268.93——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=128, Backbone=Llama2-70B2025.1268.82——————
16bitsAvg. bits=16, Group Size=-, Backbone=Llama3-8B2025.1268.64——————
SignRoundV2Avg. bits=2, Group Size=128, Backbone=Llama2-70B2025.1268.39——————
EfficientQATAvg. bits=2, Group Size=64, Backbone=Llama3-70B2025.1267.89——————
SignRoundV1 (SRV1)Avg. bits=2, Group Size=128, Backbone=Llama2-70B2025.1267.7——————
EfficientQATAvg. bits=2, Group Size=128, Backbone=Llama3-70B2025.1267.57——————
16bitsAvg. bits=16, Group Size=-, Backbone=Llama2-13B2025.1267.44——————
QuaRot-GPTQModel=Phi-3-mini, Precision=INT42024.0366.3475.3567.8872.9572.9848.1260.78
QuaRotModel=LLAMA2-7B, Quantization=4-bit (A4W4KV4), Evaluation Protocol=Zero-shot2024.0365.6476.7763.7772.1669.8740.8770.39
AQLMAvg. bits=2, Group Size=1x16, Backbone=Llama2-13B2025.1264.95——————
SignRoundV1 (SRV1)Avg. bits=2, Group Size=128, Backbone=Llama3-70B2025.1264.76——————
16bitsAvg. bits=16, Group Size=-, Backbone=Llama2-7B2025.1264.66——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=64, Backbone=Llama2-13B2025.1264.57——————
EfficientQATAvg. bits=2, Group Size=64, Backbone=Llama2-13B2025.1264.48——————
QuIP#Avg. bits=2, Group Size=-, Backbone=Llama2-13B2025.1264.44——————
AQLMAvg. bits=2, Group Size=1x16, Backbone=Llama3-8B2025.1264.1——————
SignRoundV2Avg. bits=2.5, Group Size=64, Backbone=Llama2-13B2025.1264.09——————
EfficientQATAvg. bits=2, Group Size=128, Backbone=Llama2-13B2025.1263.88——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=128, Backbone=Llama2-13B2025.1263.73——————
SignRoundV2Avg. bits=2.5, Group Size=64, Backbone=Llama3-8B2025.1263.67——————
SignRoundV2Avg. bits=2.5, Group Size=128, Backbone=Llama2-13B2025.1263.65——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=64, Backbone=Llama3-8B2025.1263.56——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=64, Backbone=Llama2-13B2025.1262.81——————
SignRoundV2Avg. bits=2.5, Group Size=128, Backbone=Llama3-8B2025.1262.76——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=128, Backbone=Llama3-8B2025.1262.6——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=128, Backbone=Llama2-13B2025.1262.34——————
AQLMAvg. bits=2, Group Size=2x8, Backbone=Llama2-13B2025.1262.22——————
SignRoundV2Avg. bits=2, Group Size=128, Backbone=Llama2-13B2025.1261.88——————
AQLMAvg. bits=2, Group Size=1x16, Backbone=Llama2-7B2025.1261.85——————
EfficientQATAvg. bits=2, Group Size=64, Backbone=Llama3-8B2025.1260.76——————
SignRoundV1 (SRV1)Avg. bits=2, Group Size=128, Backbone=Llama2-13B2025.1260.72——————
QuIP#Avg. bits=2, Group Size=-, Backbone=Llama2-7B2025.1260.61——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=128, Backbone=Llama2-7B2025.1260.44——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=64, Backbone=Llama3-8B2025.1260.38——————
SignRoundV2 (Improved Recipe)Avg. bits=2.5, Group Size=64, Backbone=Llama2-7B2025.1260.28——————
EfficientQATAvg. bits=2, Group Size=64, Backbone=Llama2-7B2025.1260.14——————
SignRoundV2Avg. bits=2.5, Group Size=64, Backbone=Llama2-7B2025.1260.08——————
OmniQAvg. bits=2, Group Size=128, Backbone=Llama3-70B2025.1260.06——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=128, Backbone=Llama3-8B2025.1259.97——————
SignRoundV2Avg. bits=2.5, Group Size=128, Backbone=Llama2-7B2025.1259.96——————
EfficientQATAvg. bits=2, Group Size=128, Backbone=Llama2-7B2025.1259.5——————
EfficientQATAvg. bits=2, Group Size=128, Backbone=Llama3-8B2025.1259.37——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=64, Backbone=Llama2-7B2025.1259.04——————
SignRoundV2Avg. bits=2, Group Size=128, Backbone=Llama3-8B2025.1259.02——————
SignRoundV2 (Improved Recipe)Avg. bits=2, Group Size=128, Backbone=Llama2-7B2025.1258.67——————
SignRoundV2Avg. bits=2, Group Size=128, Backbone=Llama2-7B2025.1257.88——————
AQLMAvg. bits=2, Group Size=2x8, Backbone=Llama2-7B2025.1257.61——————
QuaRot-RTNModel=Phi-3-mini, Precision=INT42024.0356.1268.3958.6460.665.8739.2543.99
SignRoundV1 (SRV1)Avg. bits=2, Group Size=128, Backbone=Llama3-8B2025.1255.25——————
OmniQAvg. bits=2, Group Size=128, Backbone=Llama2-70B2025.1254.87——————
DenseScale=1.3B, FLOPs=1.00, KV mem=1.002026.0754.8——————
SignRoundV1 (SRV1)Avg. bits=2, Group Size=128, Backbone=Llama2-7B2025.1254.5——————
TriRouteScale=1.3B, FLOPs=0.55, KV mem=0.402026.0754.5——————
OmniQAvg. bits=2, Group Size=128, Backbone=Llama2-13B2025.1253.56——————
Independent comboScale=1.3B, FLOPs=0.55, KV mem=0.402026.0753.5——————
OmniQAvg. bits=2, Group Size=128, Backbone=Llama3-8B2025.1252.66——————
MoD + KV-quantScale=1.3B, FLOPs=0.55, KV mem=0.402026.0752.5——————
GPTQAvg. bits=2, Group Size=128, Backbone=Llama2-13B2025.1248.29——————
DenseScale=410M, FLOPs=1.00, KV mem=1.002026.0747.2——————
TriRouteScale=410M, FLOPs=0.55, KV mem=0.402026.0747——————
OmniQAvg. bits=2, Group Size=128, Backbone=Llama2-7B2025.1246.98——————
Independent comboScale=410M, FLOPs=0.55, KV mem=0.402026.0746.2——————
MoD + KV-quantScale=410M, FLOPs=0.55, KV mem=0.402026.0745.6——————
DenseScale=160M, FLOPs=1.00, KV mem=1.002026.0742——————
TriRouteScale=160M, FLOPs=0.55, KV mem=0.402026.0741.8——————
GPTQAvg. bits=2, Group Size=128, Backbone=Llama2-7B2025.1241.56——————
Independent comboScale=160M, FLOPs=0.55, KV mem=0.402026.0740.9——————
MoD + KV-quantScale=160M, FLOPs=0.55, KV mem=0.402026.0740.3——————
AWQAvg. bits=2, Group Size=128, Backbone=Llama2-13B2025.1235.99——————
AWQAvg. bits=2, Group Size=128, Backbone=Llama2-70B2025.1235.49——————