DeepLoop debuts to make looped transformers stable and scalable, echoing 48Lx2 frontier model rumors
burny_tech · x · 2026-09-04
Yifan Zhang cites claims that some frontier models are essentially a 48-layer transformer looped twice (48L x 2), and introduces DeepLoop, a new open-source project on depth scaling for looped transformers, aimed at making the looping architecture stable and scalable.
Related event: DeepLoop: looping Transformers scale stably in depth(2 posts)→
More from Research
- Abliterated models say yes 3-4x more but miss the real FreeBSD kernel bug entirely — evilsocket · 2026-09-04
- Declarative Attention: LLMs Slash KV Cache Reads 31-52% Zero-Shot — chaumian · 2026-09-04
- CAT-Flow: Training-Free Adaptive Steps Cut Flow Matching Generation by Up to 40% — chaumian · 2026-09-04
- Absolute Zero: Self-Play RL Reasoning With Zero Human Data — cephaloform · 2026-09-04
- Go pros' decision quality and new-move invention rate jumped after AlphaGo — burny_tech · 2026-09-04
- OpenAI repo: GPT-6-Astra Lean-formalizes proof of primes within distance 186 — scaling01 · 2026-09-04