EOS

#1
by darxkies - opened

It keeps generating ignoring the EOS. Some of the quantizations of Llama 3 had that issue as well and it was fixed.

@Undi95 Needs to change field 'tokenizer.ggml.eos_token_id' from 100 to 128009 in the GGUF

Owner

It work on last version of Kobold without doing this, can you confirm?
Ooba have an issue, but I will not change if the OG repo of Llama3 don't change

Same issue with Ollama, just FYI.

Sign up or log in to comment