List your company

Common Sense Reasoning on ARC-C, ARC-E, BoolQ, HellaSwag, PIQA, WinoGrande (zero-shot)

77.67Average Accuracy (Zero-Shot)FP16 Baseline
35.092446.146257.268.2538Feb 9, 2026Feb 16, 2026Feb 24, 2026Mar 3, 2026Mar 11, 2026Mar 18, 2026Mar 26, 2026

Evaluation results

MethodLinks
FP16 BaselineModel=Llama2-70B, Precision=W16A16, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0377.6757.3480.4784.183.3282.6478.14
FP16 BaselineModel=Qwen2.5-32B, Precision=W16A16, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0377.1555.6378.0387.4384.0782.2675.45
Floating-pointAvg. Bits=16, Type=-, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0274.19——————
FP16 BaselineModel=Qwen2.5-7B, Precision=W16A16, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0373.2449.5776.0584.7178.1379.7171.27
Floating-pointAvg. Bits=16, Type=-, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0273.19——————
SpinQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0272.68——————
AWQAvg. Bits=4, Type=Static, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0272.64——————
AWQAvg. Bits=4, Type=Static, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0272.64——————
SpinQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0272.62——————
MoBiQuantAvg. Bits=3.99, Type=Elastic, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0271.84——————
OmniQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0271.68——————
OmniQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0271.66——————
MoBiQuantAvg. Bits=4.0, Type=Elastic, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0271.54——————
BF16Backbone=Llama3-8B, Quantization bits=Unquantized, Evaluation Protocol=Zero-shot2026.0271.251.7180.8981.7760.5678.8473.4
BF16Backbone=Llama3-8B, Quantization bits=Unquantized, Evaluation Protocol=Zero-shot2026.0271.251.7180.8981.7760.5678.8473.4
BF16Backbone=Qwen3-8B, Quantization bits=Unquantized, Evaluation Protocol=Zero-shot2026.0271.1955.3883.586.7357.1276.567.88
BF16Backbone=Qwen3-8B, Quantization bits=Unquantized, Evaluation Protocol=Zero-shot2026.0271.1955.3883.586.7357.1276.567.88
OBackbone=Qwen3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0270.6654.1882.6286.4555.776.3968.59
GPTQBackbone=Qwen3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0270.5654.2782.7486.7356.6276.2266.77
FP16 BaselineModel=Llama2-7B, Precision=W16A16, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0370.5446.4274.6278.0175.978.8469.46
O(R)Backbone=Qwen3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0270.553.4182.3686.0256.3476.0668.82
O(N)Backbone=Qwen3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0270.4453.7582.286.0256.3976.2268.03
AWQBackbone=Qwen3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0270.453.6782.3785.9355.9175.7968.75
RTNAvg. Bits=4, Type=Static, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0270.4——————
GPTQAvg. Bits=4, Type=Static, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0270.22——————
GPTQAvg. Bits=4, Type=Static, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0270.21——————
OBackbone=Llama3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0270.1649.1579.5981.1359.778.6772.69
AWQBackbone=Llama3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0270.0648.779.2180.859.7378.6273.32
O(R)Backbone=Llama3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0269.9749.3279.1281.0459.7978.2472.3
O(N)Backbone=Llama3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0269.948.4678.5381.046078.1873.16
GPTQBackbone=Llama3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0269.7448.2978.8980.2159.4778.2973.32
Floating-pointAvg. Bits=16, Type=-, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0269.68——————
AWQAvg. Bits=4, Type=Static, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0269.4——————
QUIPBackbone=Llama3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0269.1546.2577.3681.9358.7677.5373.09
AWQBackbone=Qwen3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0268.6550.5680.9384.4352.875.967.25
OmniQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0268.41——————
BF16Backbone=Qwen3-4B, Quantization bits=Unquantized, Evaluation Protocol=Zero-shot2026.0268.1650.3480.2685.1152.2975.365.67
BF16Backbone=Qwen3-4B, Quantization bits=Unquantized, Evaluation Protocol=Zero-shot2026.0268.1650.3480.2685.1152.2975.365.67
OBackbone=Qwen3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0268.0847.2778.4184.3954.2376.2767.9
SpinQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0268——————
O(N)Backbone=Qwen3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0267.6147.6977.6184.4653.7674.7667.35
O(R)Backbone=Qwen3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0267.5347.5977.6284.0453.8874.5967.46
SmoothQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-3-8B, Evaluation Protocol=0-shot2026.0267.47——————
SmoothQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0267.46——————
MoBiQuantAvg. Bits=3.9, Type=Elastic, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0267.22——————
OBackbone=Llama3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0266.8142.9275.4278.6955.7377.7570.32
Floating-pointAvg. Bits=16, Type=-, Backbone=LLaMA-3.2-3B, Evaluation Protocol=0-shot2026.0266.71——————
GPTQBackbone=Qwen3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0266.644.2874.6283.3353.8875.9567.56
AWQBackbone=Llama3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0266.3343.8675.2176.1555.7975.9571.03
O(R)Backbone=Llama3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0265.9941.373.4876.4556.6777.3770.64
OBackbone=Qwen3-4B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0265.9446.6578.2784.3149.8574.2162.35
O(N)Backbone=Llama3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0265.8741.8171.7176.7356.7176.8871.35
GPTQBackbone=Qwen3-4B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0265.8147.5777.982.4350.1274.4362.43
AWQBackbone=Qwen3-4B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0265.746.2977.3283.2350.3774.162.88
QUIPBackbone=Llama3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0265.6441.3871.3477.6855.6776.1271.67
GPTQAvg. Bits=4, Type=Static, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0265.54——————
SliderQuantModel=Llama2-70B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0365.2437.7164.1475.0273.0275.7965.75
RTNAvg. Bits=4, Type=Static, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0264.72——————
RTNAvg. Bits=4, Type=Static, Backbone=LLaMA-2-13B, Evaluation Protocol=0-shot2026.0264.19——————
O(R)Backbone=Qwen3-4B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0264.0944.2874.0382.9749.2872.861.17
SmoothQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-2-7B, Evaluation Protocol=0-shot2026.0263.87——————
OmniQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-3.2-3B, Evaluation Protocol=0-shot2026.0263.8——————
QUIPBackbone=Qwen3-8B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0263.345.4875.4281.9944.7269.5362.67
MoBiQuantAvg. Bits=3.94, Type=Elastic, Backbone=LLaMA-3.2-3B, Evaluation Protocol=0-shot2026.0263.2——————
O(N)Backbone=Qwen3-4B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0263.1941.8174.7983.4647.6173.1258.33
SliderQuantModel=Qwen2.5-32B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0362.7440.768.170.1462.1272.0963.3
CBQModel=Llama2-70B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0362.737.362.4572.1369.8473.2461.23
OBackbone=Qwen3-4B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0262.0341.571.4180.9545.6172.6460.06
GPTQBackbone=Qwen3-4B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0261.7639.3370.9280.9546.8170.9261.64
GPTQBackbone=Llama3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0261.5134.2166.0471.8955.2673.5668.11
AWQBackbone=Qwen3-4B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0261.1740.2768.2781.1946.0270.8760.38
QUIPBackbone=Qwen3-4B, Quantization bits=4 bits, Evaluation Protocol=Zero-shot2026.0260.8139.2571.0980.7343.6470.2459.91
CBQModel=Qwen2.5-32B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0360.7936.8663.4769.6359.4171.4463.93
QUIPBackbone=Qwen3-8B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0260.6640.2771.9377.3743.2271.3359.83
OmniQuantModel=Qwen2.5-32B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0360.3939.9361.2465.2665.4871.1659.27
OmniQuantModel=Llama2-70B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0359.9337.259.5566.366.6371.7158.17
Floating-pointAvg. Bits=16, Type=-, Backbone=LLaMA-3.2-1B, Evaluation Protocol=0-shot2026.0259.42——————
SliderQuantModel=Llama2-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0359.332.9459.6467.7762.3371.3861.72
OmniQuantAvg. Bits=4, Type=Static, Backbone=LLaMA-3.2-1B, Evaluation Protocol=0-shot2026.0256.99——————
O(R)Backbone=Qwen3-4B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0256.3937.7965.3570.5843.3967.2553.99
CBQModel=Llama2-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0356.2633.4556.461.9660.4670.1855.09
SliderQuantModel=Qwen2.5-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0356.1533.6259.362.5455.0666.9259.43
MoBiQuantAvg. Bits=3.95, Type=Elastic, Backbone=LLaMA-3.2-1B, Evaluation Protocol=0-shot2026.0255.49——————
OmniQuantModel=Llama2-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0354.9531.9154.1763.6755.9568.4455.56
O(N)Backbone=Qwen3-4B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0253.8435.9558.6869.240.6567.7450.83
SmoothQuantModel=Llama2-70B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0349.9431.2346.2157.2852.6561.3750.91
CBQModel=Qwen2.5-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0349.6931.0648.9953.6747.3162.5754.54
SmoothQuantModel=Qwen2.5-32B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0349.0932.3448.1550.5549.9161.5952.01
QUIPBackbone=Qwen3-4B, Quantization bits=3 bits, Evaluation Protocol=Zero-shot2026.0247.9124.1547.8167.8633.7262.7951.14
SmoothQuantModel=Llama2-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0345.2927.1339.7751.0741.3260.8851.54
OmniQuantModel=Qwen2.5-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0339.4824.433.8438.7833.7553.3252.8
SmoothQuantModel=Qwen2.5-7B, Precision=W4A4, Quantization Type=weight-activation, Evaluation Protocol=Zero-shot2026.0336.7326.0225.9738.8125.9250.4453.2
AQLMFinetuned=Yes, LM Eval=not reported, BPW=2.07, Details=Tables 5 & 6 of QTIP paper2026.03—32.863.774.8——65.7
Baseline (Ours)Finetuned=No, LM Eval=acc, BPW=162026.03—43.275.679.357.178.169.9
LLVQ [shape-gain, 2 bit gain]Finetuned=No, LM Eval=acc, BPW=22026.03—35.569.87349.775.266.9
LLVQ [shape-gain, 2 bit gain]Finetuned=Yes, LM Eval=acc, BPW=22026.03—39.872.975.354.177.166.3
LLVQ [spherical shaping]Finetuned=No, LM Eval=acc, BPW=22026.03—34.769.367.546.673.564.6
LLVQ [spherical shaping]Finetuned=Yes, LM Eval=acc, BPW=22026.03—40.672.970.952.575.565.1
OmniQFinetuned=Yes, LM Eval=acc, BPW=2, Details=Table 3 of Quip# paper2026.03—21.635.257.5——51.5
PV-tuningFinetuned=Yes, LM Eval=acc, BPW=2, Details=Table 8 of PV-tuning paper2026.03—38.471.2—53.57766.7