| English | 中文 |
Sequence prediction needs an explicit forecast origin, history, horizon and covariate availability boundary. It differs from treating unrelated table rows as independent samples. Read time-series study design, then the TimeXer method for the currently tracked model.
The general Transformer architecture and data guide explain attention, temporal patches, masks and alternative input structures. They complement this task-specific sequence guidance and do not move or duplicate the TimeXer capability.
The corresponding skill directory documents external Python/PyTorch usage and its reference-only status. This is not a new executable model in the tabular regression adapters.