Multiple-choice Question Answering on CommonsenseQA (dev)
76.2AccuracyRoBERTa + CSPT
Evaluation results
| Method | Links | |
|---|---|---|
| RoBERTa + CSPTpre-training=CSPT, source=leaderboard2019.10 | 76.2 | |
| ROBERTa + CSPTsource=leaderboard2019.10 | 76.2 | |
| OCN + OMCS pre-train + CN injectpre-training=OMCS, knowledge_injection=ConceptNet2019.10 | 69 | |
| OCN + OMCS pre-train + CN injectpre-training=OMCS, knowledge_source=ConceptNet2019.10 | 69 | |
| BERT + OMCS pre-trainpre-training=OMCS, source=leaderboard2019.10 | 68.8 | |
| BERT + OMCS pre-trainsource=leaderboard, pre-training=OMCS2019.10 | 68.8 | |
| OCN + CN injectionknowledge_injection=ConceptNet2019.10 | 67.3 | |
| OCN + CN injectionknowledge_source=ConceptNet2019.10 | 67.3 | |
| OCN + OMCS pre-trainpre-training=OMCS2019.10 | 65.2 | |
| OCN + OMCS pre-trainpre-training=OMCS2019.10 | 65.2 | |
| OCNdescription=baseline Option Comparison Network2019.10 | 64.1 | |
| OCN2019.10 | 64.1 | |
| OCN + ATOMIC pre-trainpre-training=ATOMIC2019.10 | 61.2 | |
| OCN + ATOMIC pre-trainpre-training=ATOMIC2019.10 | 61.2 | |
| LaMDAEvaluation Protocol=few-shot, Prompting Strategy=Chain-of-Thought (CoT), Backbone Model=LaMDA 137B2022.12 | 55.6 | |
| GPT-JEvaluation Protocol=few-shot, Prompting Strategy=Chain-of-Thought (CoT), Backbone Model=GPT-J2022.12 | 36.6 | |
| NPPromptEvaluation Protocol=zero-shot, Backbone Model=RoBERTa-large2022.12 | 34.2 | |
| GPT-JEvaluation Protocol=few-shot, Prompting Strategy=Direct, Backbone Model=GPT-J2022.12 | 20.9 |