Re-quant?

#5
by BlueNipples - opened

Just wondered if you would be requanting this now that the GGUF tokenizing in llamacpp is fixed?

You want a new GGUF quant in the gguf repo correct? I could re-upload that tonight

Which is the newer one, this or the one labelled as V1?

@concedo The V1 is named "LexiFun" it's something different. It is the first version experiment and become better in the next. This one however, is the regular Llama3-8B.

You want a new GGUF quant in the gguf repo correct? I could re-upload that tonight

Yes. There's the possibility the changed/fixed tokenization in the new llamacpp breaks old ggufs. There definitely appears to be something screwy going on when I try to run them.

For now, if anyone wants, I've created a PR with a few files re-quanted here:
https://huggingface.co/Orenguteng/Llama-3-8B-Lexi-Uncensored-GGUF/tree/refs%2Fpr%2F5

Sign up or log in to comment