This is a cleaner version of Github-code dataset, we add the following filters:

• Average line length < 100
• Alpha numeric characters fraction > 0.25
• Remove auto-generated files (keyword search)

3.39M files are removed making up 2.94% of the dataset.