codeT5+ 220M and 770M Embedding model

[feather820]

Hello, I saw in the paper that codeT5+ 770M achieved state-of-the-art results on CSN, AdvTest, and cosqa. I'm curious whether this result is based on a matching approach or an embedding approach, as I didn't find this information in the paper. If it's based on the embedding approach, would it be possible to open-source the 770M embedding model or the 220M model? Thank you very much!