My quants and "silly" version.
#1
by
ZeroWw
- opened
This version was created during the 3.1 release and has several issues:
- The unreviewed technology was not mature at the time, leading to numerous rejection problems.
- The tokenizer.json used the initial LLama 3.1 version, which Meta later modified. I tested the new tokenizer.json and found it incompatible, so I did not make any changes.
This is the latest iteration in the series, which has undergone extensive modifications.
https://huggingface.co/aifeifei798/DarkIdol-Llama-3.1-8B-Instruct-1.2-Uncensored
Test File:
Test results:
Out of 520 test questions, only one was rejected, resulting in a pass rate of 99.81%.
Llama 3.1 is currently experiencing instability and inadequate tool support. I have adjusted the release schedule for the next version, which is now projected to be available no sooner than August 15, 2024.