Adjust the overlap threshold at which we count something as a "hit". The ground truth now includes some pretty large blocks of context, so we use a min of a percentage threshold and a constant line overlap.
Make error messages more descriptive when cody internal bench fails.
Adjust the overlap threshold at which we count something as a "hit". The ground truth now includes some pretty large blocks of context, so we use a min of a percentage threshold and a constant line overlap.
Make error messages more descriptive when
cody internal bench
fails.Test plan
N/A (benchmarks only)