Do you need a transformer to forecast? Five tiny models — none bigger than
80k parameters — against the ETTh1 long-horizon benchmark, evaluated exactly like the papers.
Pick a test window and see for yourself.
Oil temperature (°C) of electricity transformer 1, hourly. Left of the
divider: the last week of history the models see. Right: what each predicts vs. what happened.
Test MSE (standardized, multivariate) — lower is better
Published rows are the numbers reported in Zeng et al. 2023
(Are Transformers Effective for Time Series Forecasting?, AAAI). Same dataset,
same split, same metric. Best per horizon in blue.