Falcon-X: A Time Series Foundation Model for Heterogeneous Multivariate Modeling

Falcon-X, a time series foundation model (TSFM), has been released to enhance heterogeneous multivariate modeling by decoupling variates from raw space into a unified latent prototype space. It introduces a Unified Prototype Diff-Attention mechanism for capturing complex interactions and employs Latent Entity Attention for efficient cross-variate interactions, achieving strong performance on GIFT-Eval and fev-bench benchmarks. This model provides a scalable solution for practitioners aiming to improve forecasting accuracy in complex multivariate environments.