chunked/nltk_chunking/model_doc_phi/chunk_3.txt

They selected high quality "textbook" data alongside with synthetically generated data for training their small sized Transformer
based model Phi-1 with 1.3B parameters.