Language Modelling On Enwiki8

Metrics

Bit per Character (BPC)

Number of params

Results

Performance results of various models on this benchmark

			Paper Title
LSTM (7 layers)	1.67	-	Generating Sequences With Recurrent Neural Networks
Hypernetworks	1.34	27M	HyperNetworks
SHA-LSTM (4 layers, h=1024, no attention head)	1.33	51M	Single Headed Attention RNN: Stop Thinking With Your Head
LN HM-LSTM	1.32	35M	Hierarchical Multiscale Recurrent Neural Networks
ByteNet	1.31	-	Neural Machine Translation in Linear Time
Recurrent Highway Networks	1.27	46M	Recurrent Highway Networks
Large FS-LSTM-4	1.25	47M	Fast-Slow Recurrent Neural Networks
Large mLSTM	1.24	46M	Multiplicative LSTM for sequence modelling
AWD-LSTM (3 layers)	1.232	47M	An Analysis of Neural Language Modeling at Multiple Scales
Cluster-Former (#C=512)	1.22	-	Cluster-Former: Clustering-based Sparse Transformer for Long-Range Dependency Encoding
LSTM	1.195	48M	Mogrifier LSTM
Mogrifier LSTM	1.146	48M	Mogrifier LSTM
64-layer Character Transformer Model	1.11	44M	Character-Level Language Modeling with Deeper Self-Attention
SHA-RNN (4 layers, h=1024, single attention head)	1.076	52M	Single Headed Attention RNN: Stop Thinking With Your Head
SHA-RNN (4 layers, h=1024, attention head per layer)	1.068	54M	Single Headed Attention RNN: Stop Thinking With Your Head
Transformer-XL (12 layers)	1.06	41M	Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
Transformer (64 layers)	1.06	235M	Character-Level Language Modeling with Deeper Self-Attention
Skip Cross-Head Transformer-XL	1.033	41M	Memory-efficient Stochastic methods for Memory-based Transformers
Transformer-XL (18 layers)	1.03	88M	Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
Transformer+SSA	1.024	-	The Information Pathways Hypothesis: Transformers are Dynamic Self-Ensembles

0 of 42 row(s) selected.

Command Palette

Language Modelling On Enwiki8

Metrics

Results