The model currently tokenizes <|begin_of_sentence|> and <|end_of_sentence|> wrong. The model has since been updated and requantizing it should probably fix it.
· Sign up or log in to comment
