bigcode/starcoder

Text Generation • Updated Oct 8 • 12.8k • • 2.81k

Note StarCoder is based on StarCoderBase and trained on additional 30B tokens of Python.

bigcode/starcoderbase

Text Generation • Updated May 11, 2023 • 728 • 394

Note StarCoderBase is a 15B parameter decoder trained on 1T tokens of code in 80+ programming languages.

bigcode/starcoderplus

Text Generation • Updated Aug 21, 2023 • 1.5k • 216

Note StarCoderPlus is based on StarCoderBase and trained on additional 600B tokens of natural text from RefinedWeb and Wikipedia.

Note The fully processed code dataset used to train StarCoder.