-
Notifications
You must be signed in to change notification settings - Fork 66
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add examples for the deita paper tasks #329
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Hi @plaguss, I realized we return the rank/scores here but I'm not sure if this is handled intuitively because normally we work with ratings where a higher rating corresponds to a higher score but in this case, it is the other way around. Right? And this might not be aligned with the RankingQuestion
in Argilla itself.
Hi @davidberenstein1957, we don't really follow the paper strictly, we decided to interpret the scores as if those were The discussion from the PR can be seen here. |
@plaguss I agree but now it seems like we attribute ratings to values that are originally rankings so they should be reversed to allow for choosing on of the preferred options. Correct? rank 1, 2 => rating 2, 1 because a lower rating should translate in a higher rank? How is this handled in the to_argilla for example? |
I understand, but for example, the from distilabel.tasks import QualityScorerTask
task = QualityScorerTask()
print(task.generate_prompt("What are the first 5 Fibonacci numbers?", ["0 1 1 2 3", "0 1 1 2 3"]).formatted_prompt) Rank the following responses provided by different AI assistants to the user’s question
according to the quality of their response. Score each response from 1 to 2, with 3
reserved for responses that are already very well written and cannot be improved further.
Your evaluation should consider factors such as helpfulness, relevance, accuracy, depth,
creativity, and level of detail of the response.
Use the following format:
[Response 1] Score:
[Response 2] Score:
...
#Question#: What are the first 5 Fibonacci numbers?
#Response List#:
[Response 1] 0 1 1 2 3
[Response 2] 0 1 1 2 3 Maybe we can discuss whether the prompts really make sense in our case and maybe rewrite them to more more "inspired" from the deita paper, but just to represent proper ratings, explaining if lower/higher is worse/better |
@plaguss Yes it makes sense. I thought we needed to rank the responses within the prompt and that they were not directly asked to score. Ignore what I've commented above. 😅 |
Description
This PR includes additional examples and adds some notes for the Deita tasks.