5fa1a76
1
2
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.
