microsoft / CodeBERT

CodeBERT
MIT License
2.09k stars 430 forks source link

How is Input Structured for Comment Generation with CodeT5 #318

Open ChunhuaLiu596 opened 2 months ago

ChunhuaLiu596 commented 2 months ago

Hello, thank you for providing the code base for CodeReviewer

I'm attempting to replicate some experimental results in Table 5 for comment generation.

I would be grateful if you could shed some light on how the input is formalized when the model is not CodeReviewer, specifically in the case of the CodeT5 model. Could you confirm whether code diff hunks are provided as input or not? If they are indeed provided, I'm curious to learn more about how the diff tags (-, +) are processed. Your timely response to this matter would be greatly appreciated. Thank you.

### Tasks