What is better 13B q4_0 or 7B q8_0 ???

[skidd-level-100]

I would like to know what kind of model is best for its parameter size,
based on file size I would guess that the 13B one is better.

please let me know

[BetaDoggo]

In terms of perplexity q4_0 is always better than the 16bit version of the lower model. There are some examples in the readme and there's a discussion with comparisons for each model.