microsoft / ProphetNet

A research project for natural language generation, containing the official implementations by MSRA NLC team.
MIT License
686 stars 109 forks source link

Why is the GENIE result in AR-diffusion very different from the original paper? Also, you come from a team. #61

Open zhaijianyang opened 1 year ago

wutong4012 commented 1 year ago

You can see the evaluation files of the two works, and you can find that the evaluation methods of the two works are actually different. This point is also highlighted in Section 4.3 of AR-DIFFUSION, which hopefully will address your concerns.