Suggestion

#3
by masure - opened

Hello,

First of all thank you very much for this model that appears to be the best for me.

I just wanted to ask if you could release an uncensored version of Chocolatine 14B Instruct DPO 1.2 model

Thank you.

This model is based on phi-3 which had little to no nsfw data in training dataset. Steering it with dpo lora won't be enough to make a decent uncensored model out of it, you have to do a full finetune with a much bigger dataset to add relevant knowledge.

w4r10ck changed discussion status to closed

Sign up or log in to comment