cascade

About cascade

Cascade is a Bittensor subnet focused on building state-of-the-art time-series foundation models through a unique competition on synthetic data. The model architecture is fixed and identical for all participants; miners compete by submitting and funding synthetic data generators that train the model. Each round, submitted generators are evaluated by training the model from a shared initialization, and the best-performing generator's model can dethrone the current 'king.' Only the crowned king's code and checkpoints are made public, while others remain private. The system emphasizes data quality as the key differentiator, with rigorous benchmarking and a transparent promotion system for top-performing models. Cascade aims to prove that superior synthetic data can drive forecasting performance, with future plans to scale model sizes and expand into multimodal forecasting.