Guide

HuggingFace PEFT: What the Docs Miss About LoRA

The default PEFT config ships with a setting that silently degrades your model at FP8. It’s not in the docs. We found it on run #34 when something looked off in the eval numbers.

In progress
Full write-up coming soon
Config diff and reproduction steps are written. Publishing soon.
Built on this research

Everything in this article is measured, not promised. See it running:

Talk to NOESIS liveSee the 33x benchmarkBack the research