Improving Factual Error Correction for Abstractive Summarization via Data Distillation and Conditional-generation Cloze
You can use main.py as a reference to use FactCloze. We propose the related models as below:
| Model | dataset | Post Alert | URL |
|---|---|---|---|
| t5-base | CNN/DM | No | https://huggingface.co/KenLee/t5_base_cnndm_sd |
| t5-base | CNN/DM | Yes | https://huggingface.co/KenLee/t5_base_cnndm_sd_pa |
| bart-large | XSum | No | https://huggingface.co/KenLee/bart_large_xsum_sd |
| bart-large | Xsum | Yes | https://huggingface.co/KenLee/bart_large_xsum_sd_pa |
Any question or problem could be sent to issue or email ~