List your company

Language Modeling on WikiText-103 1.5% corruption (test)

29.339Perplexity (PPL)Transformer-XL
28.8499232.1512135.452538.75379Mar 31, 2026

Evaluation results

MethodLinks
Transformer-XLModel size (Parameters)=Base (151M), Attention=QUEST2026.0329.339
Transformer-XLModel size (Parameters)=Base (151M), Attention=Standard2026.0329.627
TransformerModel size (Parameters)=Medium (90M), Attention=QUEST2026.0335.849
TransformerModel size (Parameters)=Medium (90M), Attention=QKNorm-DS2026.0336.018
TransformerModel size (Parameters)=Medium (90M), Attention=Standard2026.0336.234
TransformerModel size (Parameters)=Medium (90M), Attention=QNorm2026.0336.262
TransformerModel size (Parameters)=Small (44M), Attention=QUEST2026.0341.522
TransformerModel size (Parameters)=Small (44M), Attention=Standard2026.0341.566