Differences in Perplexity from the reported values

[LawrenceRLiu]

To whom it may concern,

The perplexities I am getting are different from those reported in the paper. I have attached a table (modified from table 3 of the paper), of the perplexities I am getting (red) compared to those reported.

I followed the instructions listed in INSTALL.md to setup my conda venv. The only modification I needed to get the command to run was to updating the datasets library and the corresponding lines in get_c4 as suggested here: #71

[LyoAI]

The difference seems to come from the randomness in the calibration data, which can cause slight variations in the results. However, I believe these differences are negligible as long as they are not too large.