Commonsense Reasoning on NumerSense (dev)
78AccuracyGenerated Knowledge Prompting
Evaluation results
| Method | Links | |
|---|---|---|
| Generated Knowledge PromptingInference Model=T5-11b, Knowledge Gen.=GPT-3 Knowledge2021.10 | 78 | |
| AnswersInference Model=T5-11b, Knowledge Gen.=GPT-3 Answers2021.10 | 73 | |
| Context sentencesInference Model=T5-11b, Knowledge Gen.=Context2021.10 | 70.5 | |
| Random sentencesInference Model=T5-11b, Knowledge Gen.=Random2021.10 | 68.5 | |
| Vanilla baselineInference Model=T5-11b, Knowledge Gen.=None2021.10 | 67.5 | |
| Few-shot GPT-3 Infer.2021.10 | 60.5 |