List your company

Language Modeling on WikiText-103 word-level (val)

7.533PerplexityHarmonic
6.4343213.8504121.266528.68259Oct 22, 2019Nov 27, 2020Jan 3, 2022Feb 9, 2023Mar 17, 2024Apr 23, 2025May 30, 2026

Evaluation results

MethodLinks
HarmonicSeq. len=32,7682026.057.5337.2
HarmonicSeq. len=1,0242026.057.6531.7
MambaSeq. len=32,7682026.057.69—
MambaSeq. len=1,0242026.057.722—
HarmonicSeq. len=2,0482026.057.7613.1
TransformerSeq. len=1,0242026.057.787—
HarmonicSeq. len=4,0962026.057.7924.2
HarmonicSeq. len=16,3842026.057.855.6
MambaSeq. len=2,0482026.057.868—
HarmonicSeq. len=8,1922026.057.8884.9
MambaSeq. len=4,0962026.057.895—
MambaSeq. len=16,3842026.057.963—
MambaSeq. len=8,1922026.057.988—
TransformerSeq. len=2,0482026.058.012—
TransformerSeq. len=32,7682026.058.114—
TransformerSeq. len=4,0962026.058.136—
TransformerSeq. len=8,1922026.058.292—
TransformerSeq. len=16,3842026.058.319—
Uniform Refine LSTM (UR)gate=refine, initialization=uniform2019.1033.6—
Uniform LSTM (U-)initialization=uniform2019.1033.8—
Ordered Master LSTM (OM)gate=master, initialization=ordered, activation=cumax2019.1034—
LSTM baseline2019.1034.3—
Chrono LSTMinitialization=chrono, T_max=82019.1034.3—
Chrono LSTMinitialization=chrono, T_max=112019.1034.6—
Chrono LSTMinitialization=chrono2019.1035—