It seems the quants were created before the BPE pre-tokenizer fix?

#1
by skruse - opened

I'm getting:

llm_load_vocab: missing pre-tokenizer type, using: 'default'
llm_load_vocab: ************************************
llm_load_vocab: GENERATION QUALITY WILL BE DEGRADED!
llm_load_vocab: CONSIDER REGENERATING THE MODEL
llm_load_vocab: ************************************

Would it be possible to regenerate them?

Sure thing, I'll add it to my list for today and reupload in place :)

Seems like this isn't fixed yet?

oh god thank you for the bump, it's officially in my queue sorry about that !!

updated @DDWH

Sign up or log in to comment