Worth keeping an eye on the offcial repo to see if they update to fix/use the `<|eot_id|>` token that seems to be causing lots of problems
#4
by
jukofyork
- opened
https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct/discussions/2
https://huggingface.co/meta-llama/Meta-Llama-3-70B-Instruct/discussions/12
Seems to be a particular problem for those who converted to GGUF - causing runaway self-conversations, etc.
I will modify the model when the official repo will be updated, thank you!