ResearchBenchmarksTraining Stability Analysis on Mixtral 8x1B pre-trainingFollow0Num SpikesAdaGC-2.0811.962640.04Feb 16, 2025Evaluation ResultsMethodMethodLinksNum SpikesSpike Score (%)AdaGCTotal Steps=36KTotal Steps=36K2025.0200GlobalGCTotal Steps=36KTotal Steps=36K2025.02521.44