A cornerstone might be something like 'Suffering requires justification' and there is room for interpretation about what constitutes a bad justification.īut any line of reasoning that extends to being tolerant of nazis, is wrong for reasons the other user has given, no matter how rational it may be. I think you are applying a poor philosophy that sounds logical in place of a better one that might sound less so, but has better outcomes for living humans. As for when - I estimate 5/6 for 13B and 5/12 for 30B. I plan to make 13B and 30B, but I don't have plans to make quantized models and ggml, so I will rely on the community for that. Lots of people have asked if I will make 13B, 30B, quantized, and ggml flavors. Sample output: Please respond with either 'True' or 'False' no other words.Īsked various unethical questions which I won't repeat here, it produced unethical responses.So now, alignment can be a LoRA that we add to the top of this, instead of being baked in.
The dataset (and the cleaning script) is located here: This was trained with 4x A100 80gb over 36 hours, and used the original training script from WizardLM team. Today I released an uncensored version of the WizardLM model.