This dataset is a cleaned up combination of other copyright permissive datasets. It is intended to be used as seed data in order to generate synthetic data.
It includes cleaned subsets of:
- A subset of Fineweb-edu-2 that includes CC-BY-* and .gov websites.
- A subset of open License Corpus dataset
- Youtube commons
- A subset of the starcoder data that has been clustered such that similar documents are in one sample seperated by <|endoftext|>