Trailing space after "ASSISTANT:"
#3
by
algorithm
- opened
Hello @migtissera :)
In the python example there is a trailing space after "ASSISTANT: " as seen here:
llm_prompt = f"{conversation} \nUSER: {user_input} \nASSISTANT: "
This space is missing in the prompt format in the readme. There is some evidence that this trailing space (depending on how it was trained) can affect the output.
(example: https://github.com/meta-llama/llama/issues/217#issuecomment-1774147331)
So I thought I'd add it in.
Thank you for this really nice model!
Thanks, yes I believe it affects the model generation.
migtissera
changed pull request status to
merged