Closed dyigitpolat closed 1 year ago
The smooth adaptation algorithm in WeightQuantizationTuner accepts adaptation steps with very low performance instead of shrinking the step size.
solved with commit: https://github.com/dyigitpolat/mimarsinan/commit/7fddfd4edfa6a25fa5ea0133d9e4c37105983772
The smooth adaptation algorithm in WeightQuantizationTuner accepts adaptation steps with very low performance instead of shrinking the step size.