Question Answering on ELI5
27.9B-1 ScoreKIDGPT2
Evaluation results
| Method | Links | ||
|---|---|---|---|
| KIDGPT2Backbone=GPT2, Implementation=Knowledge trie, Training Strategy=Inference-time optimization2022.04 | 27.9 | 26.6 | |
| FiD-T5Model Scale=large, Backbone=T5, Training Strategy=LM FT2022.04 | 15.7 | 18.1 | |
| FiD-T5Model Scale=base, Backbone=T5, Training Strategy=LM FT2022.04 | 15.3 | 11.3 | |
| Little KIDImplementation=naive (plain list), Training Strategy=Inference-time optimization2022.04 | 12.1 | 14.3 | |
| PPLMTraining Strategy=Inference-time optimization (LM FT / -)2022.04 | 11.9 | 15 | |
| Diverse DecodingTraining Strategy=Inference-time optimization (LM FT / -)2022.04 | 4.6 | 15.7 |