Datasets:

Modalities:
Text
Formats:
parquet
Languages:
code
Libraries:
Datasets
Dask
License:

The number of tokens

#13
by qingfuhf - opened

Hi, it seems that the number of pre-training tokens is 1T in the starcoder paper, but only 250B is available in the released dataset, did I miss anything?

Sign up or log in to comment