What types of open source code has Full Line Completion been trained on?

Question

It is stated that Full Line Completion is trained in-house using a dataset of open-source code with permissive licenses (and is not trained on any customer data). What types of open source code has Full Line Completion been trained on?

Answer

The FLCC model was trained only on code published under permissive open source licenses, specifically MIT, Apache-2.0, BSD-3-Clause, and WTFPL.