This commit is contained in:
Paul Gauthier 2024-06-03 11:23:22 -07:00
parent a0025a42d9
commit 5d08c69ba0

View file

@ -238,6 +238,7 @@ The "aider agent" internally makes multiple "attempts" at solving the problem,
but it picks and returns one single candidate solution.
Only that one candidate solution is evaluated with the acceptance tests
and contributes to the benchmark score.
Thus it is a pass@1 result.
This is contrast to a pass@N result for N>1, where N attempts are made
and all N solutions are evaluated by the acceptance tests.