List view

• Improved specificity benchmarks

  Fork and modify the code used by the MEMIT authors to run the specificity benchmark with the counterfact dataset as follows: auto-prepend the counter-factual statement with which the model has been edited (specificity+) measure specificity by KL divergence between edited and unedited model (specificity++)

  No due date

  •14/14 issues closed
  100% complete0 open14 closed