This model is designed for the Leo language and is based on `microsoft/codebert-base`.

With the code data written in the Leo language crawled from GitHub, the model is trained using self-supervised learning with triplet loss.

Anchor and positive code chunks are from the same code, while negatives are not.