·
0 commits
to a117afff5cd0aa682c1786602194feb820a7502f
since this release
- Substantially improved the performance of
convert_output_type(). Conversions are now computed per group of model/task ID combinations without first materialising ann_samples * n_output_type_idsintermediate table, reducing both runtime and memory allocation by roughly an order of magnitude for large sample tables (#282, thanks @damonbayer for reporting!).