Retains censorship from the instruct tuned model
#1
by
mpasila
- opened
Would you consider training this with the base model instead? It seems to retain the stupid "I can't do X because bla bla" text every time you do something it doesn't like.
It seems your Gemma 8.54B fine tuned model is having the same trouble of "I cant do X because bla bla bla" Gemma is a flop.
The dataset is amazing but the model isn't, nice try.
Please train it on the base model instead of the instruct model