6.5M-param NanoForecast beats 200M TimesFM on ETT after pipeline fixes cut MASE 43.8%
eulogik · hf · 2026-09-29
NanoForecast v0.5 is a 6.5M-parameter time series forecaster that rivals models 31x its size after training pipeline fixes — corrected loss-scope handling, tensor shape alignment, wider augmentation — with no architecture change, cutting MASE from 3.030 to 1.704 (-43.8%) on the same data and compute budget.
Results:
- Beats 200M-param TimesFM on all three ETT datasets and exchange rates; TimesFM keeps a clear lead on high-cardinality electricity and traffic sets;
- Wins all three ETT sets against PatchTST (15M+ params, official config);
- Trains in 12 hours on a single T4 (Colab); inference runs on CPU (Apple M4); code, checkpoints, and eval framework released under Apache 2.0.
More from Research
- Open-source AI aligns live X-ray with 3D CT at submillimeter accuracy, Nature paper — Dr_Alex_Crimi · 2026-09-29
- Tsinghua humanoid robot plays badminton with one policy from just 30 min of human motion data — ChongZzZhang · 2026-09-29
- Open-source xvr AI aligns live X-ray with 3D CT at submillimeter accuracy, published in Nature — Dr_Alex_Crimi · 2026-09-29
- Simulating Human Consciousness: Paper Maps a New Frontier for AI and Robotics — ugail · 2026-09-29
- Getting AI 'drunk' makes it more likely to break rules and spill secrets, UNSW study finds — gaganghotra_ · 2026-09-29
- 176.9B MoE squeezed to ~1.89 effective bpw: GSQ-RCO GGUFs run Coder build in 29.6GB — Loginhe · 2026-09-29