Code-Code / Clone-detection-BigCloneBench : How does a 800-length input fit into the model?

[skkuai]

Hello. Thank you for your work.

I'm testing your clone detection code. The maximum input length for the encoder in the model is 512. By the way, the block_size is 400, so the sequence length of input is 800, why does the code run without an error?

I looked at the source code document for the Roberta model, but am I missing something?

[guoday]

Two codes are fed into the model separately. So the length of input is 400 instead of 800.

[skkuai]

I got it. Thanks!