Looped transformers hint at a third scaling path for diffusion transformers
jwvdm · x · 2026-10-06
In a thread, pmpcurvo notes that Diffusion Transformers currently scale in two ways: making the network larger, or evaluating it more times during sampling (more integration steps). Looped transformers suggest a third option: reusing the same parameters repeatedly, increasing executed depth without increasing model size — a potentially new scaling direction for diffusion models.
More from Research
- Uni-LaDiR unifies image, text and 3D reasoning via latent diffusion thoughts — Lianhuiq · 2026-10-06
- Critic Calls Out 'Research Taste Doubles Every 3 Months' Eval as Merely Metric Optimization — dhadfieldmenell · 2026-10-06
- Researchers push to cap conference submissions at 1–3 per author to kill paper slop — moarbugs · 2026-10-06
- MoSE3 estimates per-pixel SE(3) transforms from video, NeurIPS Spotlight — du_yilun · 2026-10-06
- SimEX lets coding agents master real robot skills via sim-integrated autoresearch — yifengzhu_ut · 2026-10-06
- PFN features power tabular anomaly detection: ZEN tops ADBench without fine-tuning — Maximilian Bershtman · 2026-10-06