List your company

Zero-shot Question Answering on Evaluation Suite (WinoGrande, PiQA, HellaSwag, ARC-E, ARC-C)

69.39WinoGrande AccuracyPocketLLM
51.314856.007460.765.3926Nov 19, 2025

Evaluation results

MethodLinks
PocketLLMCompress Ratio=2x~8x, Avg bits=3.98, Fine-tuned=true2025.1169.3978.5457.4576.1843.1764.95
GPTQCompress Ratio=2x~8x, Avg bits=4.00, Fine-tuned=true2025.1168.1976.6155.4466.236.7760.64
PocketLLMCompress Ratio=2x~8x, Avg bits=3.98, Fine-tuned=false2025.1168.1976.9355.0173.9941.3863.1
VPTQCompress Ratio=~ 10x, Avg bits=3.01, Fine-tuned=true2025.116877.35669.139.361.72
GPTVQCompress Ratio=~ 10x, Avg bits=3.13, Fine-tuned=true2025.1167.6477.64—72.7343.69—
PocketLLMCompress Ratio=~ 10x, Avg bits=2.98, Fine-tuned=true2025.1167.478.1357.1774.1243.5264.07
AQLMCompress Ratio=2x~8x, Avg bits=4.04, Fine-tuned=true2025.1167.3278.2455.9970.1641.0462.55
Llama 2-7BCompress Ratio=-, Avg bits=322025.1167.2578.4556.6976.0143.0364.29
PocketLLMCompress Ratio=~ 16x, Avg bits=2.02, Fine-tuned=true2025.1167.2576.7153.2469.0736.7760.61
VPTQCompress Ratio=2x~8x, Avg bits=4.01, Fine-tuned=true2025.1167.178.1556939.761.98
QTIPCompress Ratio=2x~8x, Avg bits=4.01, Fine-tuned=true2025.1167.178.4—68.940—
SpQRCompress Ratio=2x~8x, Avg bits=3.98, Fine-tuned=true2025.1166.9378.3556.169.1139.6862.17
AQLMCompress Ratio=~ 10x, Avg bits=3.04, Fine-tuned=true2025.1166.9376.8854.1268.0638.460.88
QTIPCompress Ratio=~ 10x, Avg bits=3.01, Fine-tuned=true2025.1166.978.1—68.138.9—
QuIP#Compress Ratio=2x~8x, Avg bits=4.02, Fine-tuned=true2025.1166.8577.9155.7868.0639.6861.66
WandaCompress Ratio=2x~8x, Avg bits=8.00, Fine-tuned=false2025.1166.61—46.4569.1535.84—
SoBPCompress Ratio=2x~8x, Avg bits=11.20, Fine-tuned=false2025.1166.2273.5—59.8137.63—
BolacoCompress Ratio=2x~8x, Avg bits=11.20, Fine-tuned=true2025.1165.6775.19—69.1538.74—
AQLMCompress Ratio=~ 16x, Avg bits=2.02, Fine-tuned=true2025.1165.6774.7649.5563.6832.7657.28
BolacoCompress Ratio=2x~8x, Avg bits=11.20, Fine-tuned=false2025.1164.871.44—66.4634.39—
QTIPCompress Ratio=~ 16x, Avg bits=2.00, Fine-tuned=true2025.1164.775.9—67.2535.7—
VPTQCompress Ratio=~ 16x, Avg bits=2.02, Fine-tuned=true2025.1164.3375.1952.0863.835.2458.13
LLM-PrunerCompress Ratio=2x~8x, Avg bits=11.20, Fine-tuned=true2025.1164.2576.01—66.6237.2—
DuQuantCompress Ratio=2x~8x, Avg bits=4.00, Fine-tuned=true2025.1163.9375.68—5037.46—
SpQRCompress Ratio=~ 10x, Avg bits=2.98, Fine-tuned=true2025.1163.5474.8151.8567.4237.7159.07
PocketLLMCompress Ratio=~ 10x, Avg bits=2.98, Fine-tuned=false2025.1163.5475.4151.6769.9937.0359.53
SoLACompress Ratio=2x~8x, Avg bits=11.20, Fine-tuned=false2025.1162.8369.97—59.1332.68—
QuIP#Compress Ratio=~ 16x, Avg bits=2.02, Fine-tuned=true2025.1162.4371.3842.9455.5628.8452.23
DuQuantCompress Ratio=2x~8x, Avg bits=4.00, Fine-tuned=false2025.1162.1275.24—51.8936.77—
GPTVQCompress Ratio=~ 16x, Avg bits=2.13, Fine-tuned=true2025.1160.9370.73—58.0831.48—
PocketLLMCompress Ratio=~ 20x, Avg bits=1.60, Fine-tuned=true2025.1160.373.547.6363.5832.2555.45
GPTQCompress Ratio=~ 10x, Avg bits=3.00, Fine-tuned=true2025.1159.1971.4945.2158.4631.0653.08
LLM-PrunerCompress Ratio=2x~8x, Avg bits=11.20, Fine-tuned=false2025.1158.7271.93—61.4133.96—
PocketLLMCompress Ratio=~ 16x, Avg bits=2.02, Fine-tuned=false2025.1158.7270.9544.0258.1630.3852.45
PocketLLMCompress Ratio=~ 20x, Avg bits=1.60, Fine-tuned=false2025.1152.0154.0827.233.0423.2137.91