← All threads

Optimizer Unification and Efficiency

Unifying optimizer variants (e.g., Muon) through composed matrix normalizations to reduce pretraining token use.

3 papers

Where this stands

The written synthesis of this thread is for subscribers. Subscribe.