Jun 20, 2023
Microsoft researchers introduce a 1.3-billion-parameter code model trained on selected web data and synthetic textbooks.
The paper shows how training-data quality can make small models more capable.
Explore on Humanity Unfolds