See issue title. There are very likely more optimizations that can be done to improve its speed. I first noticed the issue in automated testing with GitHub Actions, but it could have an affect on actual models as well.