Llama3 8B Instruct doesn't generate EOS nor EOT tokens consistently. #8176
Unanswered
AymenSekhri
asked this question in
Q&A
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
I am trying to use simple example on Llama3 8B instruct (I tried several variations of Llama3 8B instruct model) but it fails to stop talking, AKA it doesn't generate EOS nor EOT tokens!
According to Meta's documentation the format for the prompt is something like:
The <|begin_of_text|> token should be included by llama_tokenize function with add_special = true.
The output starts good, but it doesn't know when to stop, e.g. the previous prompt will generate
It only stops when it hits the max output length! However some prompts (same format) actually generates a proper text that ends with <|eot_id|> then EOS, but it is not consistent. I also tried ending the prompts lines with "\r\n\n" instead of "\n\n" but it doesn't work for all prompts again.
What am doing wrong?
Beta Was this translation helpful? Give feedback.
All reactions