It generates <0x0A> instead of new line
#1
by
Hoioi
- opened
I'm not sure if there's something wrong with the original model or the quantized version, but it generates <0x0A> instead of a going to the new line.
OK thanks. I'm also not sure if this is because of the original model, or a bug in the code I need to use to make it.
Because the original model doesn't include a tokenizer.model, I need to use a llama.cpp PR that can make GGUF from tokenizer.json. And it currently seems to have this bug.
I will see if I can find a tokenizer.model for this model and do it that way instead.
I've re-created the GGUFs for this model and v1 and newlines are now generated correctly.
Thank you so much for your fast support!