Codspeed benchmarks sometimes run on different devices, that was fine when the runtimes were quite long, but for some of the quicker benchmarks it's tipping over the 3% tolerance for warning/acknowledging.
Additionally, the simulated annealing benchmarks were removed because they were too noisy, but that means its been extra work to spot that #204 is going to introduce a fairly significant performance regression for randomised subsystem descent.
Codspeed benchmarks sometimes run on different devices, that was fine when the runtimes were quite long, but for some of the quicker benchmarks it's tipping over the 3% tolerance for warning/acknowledging.
Additionally, the simulated annealing benchmarks were removed because they were too noisy, but that means its been extra work to spot that #204 is going to introduce a fairly significant performance regression for randomised subsystem descent.