TAPS scheduler adaptively scales recurrent updates in looped transformers, up to 1.56x speedup
Boyuan Wang · hf · 2026-10-01
Recurrent reasoning models apply updates at a fixed scale — conservative during persistent progress, aggressive during fluctuation. TAPS (Trajectory Adaptive Progress-Fluctuation Scheduler) decomposes loss sensitivity to update scale into progress and fluctuation terms and adapts step size online, with theoretical guarantees. Empirically it improves terminal accuracy on structured reasoning without retraining, and yields up to 1.56x wall-clock speedup at matched accuracy when incorporated into training, across diverse recurrent architectures.
More from Research
- Intent Lab's review agent catches 93% of 284 real bugs at 63% lower cost — jiayq · 2026-10-01
- RICE turns any LLM into a dense retriever with zero training — CShorten30 · 2026-10-01
- Functional ultrasound imaging: a computational deep dive into the new imaging modality — patrickmineault · 2026-10-01
- Nature editorial: AI medical devices need rigorous real-world testing — EricTopol · 2026-10-01
- Neuroscientists Find Surprisingly Complex Spiral Vortex Waves Traveling Across the Cortex — JosephJacks_ · 2026-10-01
- Hutter Prize sees biggest year in 20-year history: three winning entries cut size ~10% in one month — mhutter42 · 2026-10-01