Cloze-style word prediction on LAMBADA medium model (test)
61.3AccuracyGPT-2 + Coref
Evaluation results
| Method | Links | ||
|---|---|---|---|
| GPT-2 + CorefModel scale=Medium, Coreference augmentation=Yes, Evaluation protocol=Finetuned2022.08 | 61.3 | 4.58 | |
| GPT-2Model scale=Medium, Coreference augmentation=No, Evaluation protocol=Finetuned2022.08 | 59.1 | 4.59 | |
| GPT-2 (Zero-shot)Model scale=Medium, Evaluation protocol=Zero-shot, Casing=Preserved2022.08 | 55.5 | 15.6 |