Character-Level Language Modeling with Deeper Self-Attention
LSTMs and other RNN variants have shown strong performance on character-level language modeling. These models are typically trained using truncated backpropagation through time, and it is common to assume that their success stems from their ability to remember long-term contexts. In this paper, we show that a deep (64-layer) transformer model with fixed context outperforms RNN variants by a large margin, achieving state of the art on two popular benchmarks: 1.13 bits per character on text8 and 1.06 on enwik8. To get good results at this depth, we show that it is important to add auxiliary losses, both at intermediate network layers and intermediate sequence positions.
PDF Abstract
Tasks
Datasets
| Task | Dataset | Model | Metric Name | Metric Value | Global Rank | Benchmark |
|---|---|---|---|---|---|---|
| Language Modelling | enwik8 | Transformer (64 layers) | Bit per Character (BPC) | 1.06 | # 25 | |
| Number of params | 235M | # 5 | ||||
| Language Modelling | Hutter Prize | 64-layer Character Transformer Model | Bit per Character (BPC) | 1.06 | # 8 | |
| Number of params | 235M | # 3 | ||||
| Language Modelling | Hutter Prize | 12-layer Character Transformer Model | Bit per Character (BPC) | 1.11 | # 11 | |
| Number of params | 44M | # 13 |
