Character-level Prediction on PTB (test)
1.045BPC (Test)CKCNN-Big
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| CKCNN-BigNumber of Parameters=1M2021.02 | 1.045 | — | — | |
| TrellisNetSize=13.4M2018.10 | 1.158 | — | — | |
| TrellisNetNumber of Parameters=8M2021.02 | 1.158 | — | — | |
| AWD-LSTM+PDR#Params=13.8M2018.08 | 1.169 | — | — | |
| AWD-LSTM#Params=13.8M2018.08 | 1.175 | — | — | |
| AWD-LSTMSize=13.8M2018.10 | 1.175 | — | — | |
| Quasi-RNNSize=13.8M2018.10 | 1.187 | — | — | |
| Fast-Slow-LSTM-2Size=7.2M2018.10 | 1.19 | — | — | |
| FS-LSTM-4#Params=6.5M2018.08 | 1.193 | — | — | |
| NAS Cell#Params=16.3M2018.08 | 1.214 | — | — | |
| NAS CellSize=16.3M2018.10 | 1.214 | — | — | |
| HyperLSTM#Params=14.4M2018.08 | 1.219 | — | — | |
| HyperLSTMSize=14.4M2018.10 | 1.219 | — | — | |
| Independently RNNSize=12.0M2018.10 | 1.23 | — | — | |
| HM-LSTM2018.08 | 1.24 | — | — | |
| LEM# units=1024, # params=5M2021.10 | 1.25 | — | — | |
| Zoneout LSTM2018.08 | 1.27 | — | — | |
| Generic TCNSize=3.0M2018.10 | 1.31 | — | — | |
| TCNNumber of Parameters=70K2021.02 | 1.31 | — | — | |
| LSTM# units=1000, # params=5M2021.10 | 1.36 | — | — | |
| LSTMNumber of Parameters=70K2021.02 | 1.36 | — | — | |
| GRU# units=1024, # params=3M2021.10 | 1.37 | — | — | |
| GRUNumber of Parameters=70K2021.02 | 1.37 | — | — | |
| LSTMBPTT length (T)=1502023.04 | 1.41 | — | — | |
| Lipschitz RNN# units=764, # params=1.3M2021.10 | 1.42 | — | — | |
| LSTMBPTT length (T)=3002023.04 | 1.43 | — | — | |
| coRNN# units=1024, # params=2.3M2021.10 | 1.46 | — | — | |
| asRNNBPTT length (T)=1502023.04 | 1.46 | — | — | |
| nnRNN# units=1437, # params=1.3M2021.10 | 1.47 | — | — | |
| nnRNNBPTT length (T)=1502023.04 | 1.47 | — | — | |
| asRNNBPTT length (T)=3002023.04 | 1.49 | — | — | |
| nnRNNBPTT length (T)=3002023.04 | 1.49 | — | — | |
| expRNNBPTT length (T)=1502023.04 | 1.49 | — | — | |
| expRNN# units=1437, # params=1.3M2021.10 | 1.51 | — | — | |
| expRNNBPTT length (T)=3002023.04 | 1.52 | — | — | |
| anti.sym RNN# units=1437, # params=1.3M2021.10 | 1.6 | — | — | |
| EURNNBPTT length (T)=1502023.04 | 1.61 | — | — | |
| EURNNBPTT length (T)=3002023.04 | 1.62 | — | — | |
| RNN-orthBPTT length (T)=1502023.04 | 1.62 | — | — | |
| RNN-orthBPTT length (T)=3002023.04 | 1.66 | — | — | |
| RNNBPTT length (T)=1502023.04 | 2.89 | — | — | |
| RNNBPTT length (T)=3002023.04 | 2.9 | — | — | |
| Antisymmetric RNN# params=≈1.32M2020.06 | — | 1.6 | 1.64 | |
| EURNN# params=≈1.32M2020.06 | — | 1.61 | 1.62 | |
| Exponential RNN# params=≈1.32M2020.06 | — | 1.49 | 1.52 | |
| Lipschitz RNN using Euler# params=≈1.32M2020.06 | — | 1.43 | 1.46 | |
| Non-normal RNN# params=≈1.32M2020.06 | — | 1.47 | 1.49 | |
| RNN baseline# params=≈1.32M2020.06 | — | 2.89 | 2.9 | |
| RNN-orth# params=≈1.32M2020.06 | — | 1.62 | 1.66 |