Language Modeling on LM (val)
10.18Validation PPLRejection Sampling
Evaluation results
| Method | Links | |
|---|---|---|
| Rejection SamplingBackbone=1.3B LM, Decoding Speed=4x slow2022.02 | 10.18 | |
| DAPT (nontoxic)Backbone=1.3B LM2022.02 | 10.4 | |
| SGEAT (heuristic)Backbone=1.3B LM, Prompt Design=heuristic2022.02 | 11.14 | |
| SGEAT (augmented)Backbone=1.3B LM, Prompt Design=augmented2022.02 | 11.19 | |
| SGEAT + Rejection SamplingBackbone=1.3B LM2022.02 | 11.19 | |
| SGEAT (standard)Backbone=1.3B LM, Prompt Design=standard2022.02 | 11.22 | |
| Jigsaw (nontoxic)Backbone=1.3B LM2022.02 | 11.51 | |
| DEXPERTSBackbone=1.3B LM, Decoding Speed=3x slow2022.02 | 19.87 | |
| SGEAT + DEXPERTSBackbone=1.3B LM2022.02 | 20.21 |