List your company

Language Modeling on WikiText-103 2.5% corruption (test)

33.008Perplexity (PPL)Transformer-XL
32.4731236.0835639.69443.30444Mar 31, 2026

Evaluation results

MethodLinks
Transformer-XLModel size (Parameters)=Base (151M), Attention=QUEST2026.0333.008
Transformer-XLModel size (Parameters)=Base (151M), Attention=Standard2026.0333.386
TransformerModel size (Parameters)=Medium (90M), Attention=QUEST2026.0340.499
TransformerModel size (Parameters)=Medium (90M), Attention=QKNorm-DS2026.0340.715
TransformerModel size (Parameters)=Medium (90M), Attention=QNorm2026.0340.853
TransformerModel size (Parameters)=Medium (90M), Attention=Standard2026.0340.866
TransformerModel size (Parameters)=Small (44M), Attention=QUEST2026.0346.299
TransformerModel size (Parameters)=Small (44M), Attention=Standard2026.0346.38