This model is a bit too crazy and doesn't listen

#1
by xiaoyangtuo - opened

Both the prompt words and the reminders during the conversation are useless. The language used is too vulgar and I can't stand it

Owner

It’s because the data i trained it with is a bit crazy. I am aware of it and currently in the process of training a more coherent yet uncensored model using full finetuning instead of LoRa adapters.

While I understand the training corpus is somewhat coarse, the core problem is instruction adherence — or rather, the lack thereof. Its ability to follow constraints is severely lacking. I have not encountered an LLM this unresponsive to prompting; standard prompt-level guardrails are simply ineffective.

according to gptzero your last message is 100% AI generated. My LLM models are not intended to be used by users located in countries outside NATO member states.

Sign up or log in to comment