List your company
ResearchFormerly WizWand

Language Modeling on WikiText-103 v1 (test)

10.48PerplexityClean
10.19612.11314.0315.947May 12, 2026
Updated 5mo ago

Evaluation results

MethodLinks
CleanBackbone=OLMoE, Temperature=25°C2026.0510.48
CleanBackbone=OLMoE, Temperature=80°C2026.0510.48
CleanBackbone=DeepSeek, Temperature=25°C2026.0510.64
CleanBackbone=DeepSeek, Temperature=80°C2026.0510.64
ROMERBackbone=OLMoE, Temperature=25°C2026.0510.82
CleanBackbone=Qwen1.5, Temperature=25°C2026.0510.85
CleanBackbone=Qwen1.5, Temperature=80°C2026.0510.85
Bit-slicingBackbone=OLMoE, Temperature=25°C2026.0510.86
ROMERBackbone=DeepSeek, Temperature=25°C2026.0510.86
ROMERBackbone=Qwen1.5, Temperature=25°C2026.0510.9
Bit-slicingBackbone=Qwen1.5, Temperature=25°C2026.0510.93
Weight-remappingBackbone=Qwen1.5, Temperature=25°C2026.0510.93
Weight-remappingBackbone=DeepSeek, Temperature=25°C2026.0510.95
Weight-remappingBackbone=OLMoE, Temperature=25°C2026.0510.96
k-b calibrationBackbone=Qwen1.5, Temperature=25°C2026.0510.97
VanillaBackbone=Qwen1.5, Temperature=25°C2026.0511
AverageBackbone=OLMoE, Temperature=25°C2026.0511.07
AverageBackbone=Qwen1.5, Temperature=25°C2026.0511.09
AverageBackbone=DeepSeek, Temperature=25°C2026.0511.13
k-b calibrationBackbone=DeepSeek, Temperature=25°C2026.0511.13
ROMERBackbone=Qwen1.5, Temperature=80°C2026.0511.16
VanillaBackbone=DeepSeek, Temperature=25°C2026.0511.25
k-b calibrationBackbone=OLMoE, Temperature=25°C2026.0511.25
Weight-remappingBackbone=Qwen1.5, Temperature=80°C2026.0511.29
ROMERBackbone=DeepSeek, Temperature=80°C2026.0511.31
ROMERBackbone=OLMoE, Temperature=80°C2026.0511.43
VanillaBackbone=OLMoE, Temperature=25°C2026.0511.45
k-b calibrationBackbone=Qwen1.5, Temperature=80°C2026.0511.55
Weight-remappingBackbone=DeepSeek, Temperature=80°C2026.0511.58
Bit-slicingBackbone=Qwen1.5, Temperature=80°C2026.0511.59
VanillaBackbone=Qwen1.5, Temperature=80°C2026.0511.72
Bit-slicingBackbone=DeepSeek, Temperature=80°C2026.0511.78
Weight-remappingBackbone=OLMoE, Temperature=80°C2026.0511.82
AverageBackbone=Qwen1.5, Temperature=80°C2026.0511.95
k-b calibrationBackbone=DeepSeek, Temperature=80°C2026.0512.13
AverageBackbone=DeepSeek, Temperature=80°C2026.0512.47
VanillaBackbone=DeepSeek, Temperature=80°C2026.0512.49
Bit-slicingBackbone=DeepSeek, Temperature=25°C2026.0512.54
k-b calibrationBackbone=OLMoE, Temperature=80°C2026.0512.6
AverageBackbone=OLMoE, Temperature=80°C2026.0512.9
VanillaBackbone=OLMoE, Temperature=80°C2026.0513.13
Bit-slicingBackbone=OLMoE, Temperature=80°C2026.0514.62
CleanBackbone=Qwen3, Temperature=25°C2026.0514.87
CleanBackbone=Qwen3, Temperature=80°C2026.0514.87
ROMERBackbone=Qwen3, Temperature=25°C2026.0514.97
Weight-remappingBackbone=Qwen3, Temperature=25°C2026.0515.02
k-b calibrationBackbone=Qwen3, Temperature=25°C2026.0515.11
VanillaBackbone=Qwen3, Temperature=25°C2026.0515.16
ROMERBackbone=Qwen3, Temperature=80°C2026.0515.32
AverageBackbone=Qwen3, Temperature=80°C2026.0515.44
Weight-remappingBackbone=Qwen3, Temperature=80°C2026.0515.51
AverageBackbone=Qwen3, Temperature=25°C2026.0515.64
k-b calibrationBackbone=Qwen3, Temperature=80°C2026.0515.89
VanillaBackbone=Qwen3, Temperature=80°C2026.0516.14
Bit-slicingBackbone=Qwen3, Temperature=25°C2026.0516.52
Bit-slicingBackbone=Qwen3, Temperature=80°C2026.0517.58