Datasets:
Discrepancy in Reported vs. Observed Token Counts for Smol and Full Datasets
#10
by
shaharukh-khan-ola
- opened
Hi @anton-l ,
We have downloaded both the smol and full datasets and have encountered a discrepancy between the number of tokens we've observed and the figures stated in the associated paper.
Our word and token counts are as follows:
| Dataset Name | Observed Number of Words | Observed Number of Tokens (1.5 t.w.r.) |
|---|---|---|
| smol | ~110B | ~165B |
| Full | ~160B | ~240B |
Could you please provide some clarification on this difference? We want to ensure we are processing the datasets correctly and have the accurate baseline figures for our work.
Thank you for your time and assistance.