Help with convert-hf-to-gguf-update.py
I created Meta-Llama-3-8B-Instruct-f32-Q8_0.gguf to test it... but the tokenizer is messed up by the previous version...
how do I use convert-hf-to-gguf-update.py to fix it?
You need to convert again from the beginning.
You need to use the ...update.py script as explained in the Warning box, in the card, to get the llama-bpe tokenizer config files, then replace the ones in your original HF-Model files folder.
HF model → outfile BF16 GGUF → Quant
I recommend just using the lossless script from this repo, during the model download, replace the config files and it should work.
Adding myself here so I can reply more easily.
Currently you should be good to use the default configs with the recent models, since they should all have them correctly by now.