More info
Can you please provide more info about this model?
Is the base model llama 13B?
Was it fine-tuned on the entire dolphin dataset?
How good is it?
Thank you
I've not yet published
this is now published.
the base model is llama 13b.
I uncensored, filtered, cleaned the dataset. But yes it's the full release not a checkpoint or sample.
It is very good.
this is now published.
the base model is llama 13b.
I uncensored, filtered, cleaned the dataset. But yes it's the full release not a checkpoint or sample.
It is very good.
Thanks for the heads up.
Would you mind sharing which package you used?
Did you use huggingface stftrainer?
this is now published.
the base model is llama 13b.
I uncensored, filtered, cleaned the dataset. But yes it's the full release not a checkpoint or sample.
It is very good.
You mean you removed refusals for bigotry in order to introduce your own putrid political bias. Be honest, at least. Just say "I delete minorities because I don't think they deserve representation and want to introduce toxic bias".
Again @mdegans continues his obsessed campaign of personal harassment against me.
Get a life.
The time for words is through. I am proceeding with legal action.
You don't have grounds. You're a limited purpose public figure so a defamation action is right out. There's nothing else you can do. Criticism of your dataset filtering is protected speech. Something you should be in favor of, no? Want me to go away? Stop removing minorities from datasets and calling it "Uncensored" or "unfiltered". It's not. You are literally filtering a dataset for ideas you don't like, like tolerance, feminism, consent, and so on. You filter out any mention of LGBT people leaving behind the harmful biases of the base model. At a time when hate speech is on the rise, you openly bragged about how horrible the content was that your model generated.
