There is a weird yet significant memory usage increase in many seemingly unrelated scikit-learn estimators that happened on 2023-03-22, e.g.:

• https://scikit-learn.org/scikit-learn-benchmarks/#neighbors.KNeighborsClassifierBenchmark.peakmem_fit?p-algorithm='brute'&p-dimension='low'&p-n_jobs=4&commits=e3d1f9ac-18af5508

• https://scikit-learn.org/scikit-learn-benchmarks/#linear_model.LogisticRegressionBenchmark.peakmem_predict?p-representation='sparse'&p-solver='saga'&p-n_jobs=1&commits=e3d1f9ac-18af5508

It would be necessary to inspect what change between e3d1f9a and 18af550 could be responsible for this.