This is cool but would be even cooler if it could do code similarity search. The cost would probably be prohibitively high though unless you only did like 10000 repos. https://github.com/microsoft/CodeBERT/tree/master/UniXcoder#...