Bias?
#2
by
kmmmmmmmm
- opened
My previous post on this subject seems to have vaporized.
This is an almost unusable model due to hard core insistence on its "opinions" which leaks into all areas, not just political matters. It produces some hilarious results if asked to correct it's output on matters controversial or trivial, even with RAG methods.
Is this Llama 2 behavior in general? If so, I would be interested if budecosystem could tune the toxic nature out of this abomination.
@kmmmmmmmm This is a quantized version of the genz-13b-v2. Quantized models can have an impact on the performance. Please try the genz-13b-v2 for the full performance of the model.
ok