Question about discrepancy between Table 1 token counts and HF released dataset
#2
by
Elio040208 - opened
Hi, thanks for open-sourcing this dataset!
I have a question regarding Table 1 in the paper. In the table, the total token count for L4 + L5 is reported as 906.5B tokens. However, the released version on Hugging Face (L3–L5) appears to contain only 496B tokens.
I would like to clarify:
Is the HF release a subset of the data described in the paper?
If so, what were the filtering or selection criteria used to construct the released subset?
Are there any plans to release the remaining portion of the dataset in the future?
Thanks in advance for your clarification!