Do you plan to provide a GGUF update with the BPE tokenizer fix?
You may be aware that there was an issue with the tokenizer for llama-3 and GGUF files:
do you plan to provide updated GGUF for this model, or the original 32b model so that we can regenerate the GGUF with the fix ourselves? Thanks for this awesome extension!
Hi,
Yes, but there are keep adding more improvements to the Llama-3 tokenizer (regex/patterns). Since there are a lot of models which are in the pipe, I am waiting for it to settle and be sure there are no immediate improvements required. (a couple of us already used and evaluated these GGUFs, in some cases the output is identical for before and after that fix. Some edge-cases, some conversations with history, etc.)
So the models are very useable, but for sure I am planning to re-do them as soon as I see no Llama-3 related PRs for 24 hours