Richer Feature Representations via Concatenation during Fixed-Time Pretraining - View it on GitHub