Which dataset used for quantization?
#1
by
Yhyu13
- opened
Hi,
The author has mentioned this in the readme:
Please do not use wikitext for quantization calibration because all wikitext have been re-aligned on synthetic dataset, and its distribution differs significantly from the original wikitext.
I wonder which dataset has been used for quantization ?
Thanks!
exllamav2 uses its own built-in dataset by default for quantization. It's a mix of lots of different datasets (wiki, RP, coding, languages, etc.)