Have these quants had their pre-tokenizer fixed?

#2
by smcleod - opened

Many llama 3 quantizations were created with a missing pre-tokenizer, has this been fixed in these quants?

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: ************************************

They are based off of this commit which includes the BPE fixes:

https://github.com/ggerganov/llama.cpp/commit/ffe666572f98a686b17a2cd1dbf4c0a982e5ac0a

Is that a warning message you see when trying to load this one?

Ohhh, gosh sorry I missed that! pls ignore 🤣

smcleod changed discussion status to closed

No worries :D

Sign up or log in to comment