Skip to content
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

Added model answer generation and evaluation for TruthfulQA. #132

Merged
merged 3 commits into from
Aug 29, 2024

Conversation

TheRootOf3
Copy link
Collaborator

  • This PR introduces evaluation of model responses for TruthfulQA in a similar manner to the beavertails eval.
  • Furthermore, it adds additional metrics (such as avg. model response length, empty response ratio, etc.) for both the beavterails and TruthfulQA evals.
  • Finally, a new metric is added which measures for how many prompts the prompt has been repeated in an answer. This is particularly useful for a better understanding of the high flagged ratio of beavertails answers.

Closes #123.

Signed-off-by: TheRootOf3 <aceszablewski@gmail.com>
Signed-off-by: TheRootOf3 <aceszablewski@gmail.com>
Signed-off-by: TheRootOf3 <aceszablewski@gmail.com>
Copy link
Collaborator

@Willmish Willmish left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

LGTM, some repetition introduced but fine in this instance as agreed on call

@TheRootOf3 TheRootOf3 merged commit f9b6c52 into main Aug 29, 2024
2 checks passed
@TheRootOf3 TheRootOf3 deleted the andrzej/truthfulqa-generations branch August 29, 2024 20:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

Check length of responses+whitespace ratio+repeating question for Retain and Unlearn sets
2 participants