Starting from scratch: Training a 30M Topological Transformer