looped-transformers

A theoretical reconstruction of the Claude Mythos architecture, built from first principles using the available research literature.

LoopFormer is an elastic-depth looped Transformer trained on variable-length trajectories, using time/step-size conditioning and a shortcut-consistency objective to enable budget-conditioned language modeling and latent reasoning without retraining.

An open, visual atlas of looped transformers, recurrent computation, adaptive depth, and latent reasoning.

A bilingual atlas of looped and recursive Transformers, latent reasoning, and test-time compute. 持续更新的 Loop Transformer 双语研究图谱。

Experimental implementation of "Looped Transformers are Better at Learning Learning Algorithms" showing superior performance with 12x fewer parameters. Includes complete environment setup, pre-trained weights, and extensive experiments comparing Looped TFs vs traditional Transformers for In-Context Learning.

Awesome list of papers, code, models and blogs on Looped / recurrent-depth / weight-tied Transformers — depth as a third scaling axis

Evaluator probes and branch-selection scaffolding for Ouro-RLTT loop states.

Jacobian stability bounds for a looped Transformer block with MSA, FFN and RMSNorm (AINL 2026)

Lightweight pairwise evaluator for relational signals in Ouro-2.6B-Thinking loop-state trajectories.

A comprehensive experimental lab for Looped Transformers from scratch. Includes an automated ablation pipeline (ExperimentTable) and a real-time monitoring dashboard.

Lifetime meta-learning on Ouro-RLTT looped transformers: the O1 oracle-reachability program (sealed v1.5.3->v2.1 + B200 runner), M+N backbone-training and branch-training reports. Companion to Hidden-State-Evaluator (RPE) and Branching-Looped-Transformer (OPI).

An implementation of the leworldmodel using looped transformer

Deployment-blocking bugs, a Looped-Transformer memory bound, and a chunked-prefill fix for running Nanbeige4.2-3B on Apple Silicon (MPS)

awesome-looped-transformers

Benchmark and evaluation ByteDance Ouro model based on Looped Language Models on several reasoning tasks.

J-Lens (Jacobian lens) versus the raw logit lens in the looped Ouro-2.6B: discovery, independent refits, frozen 160-item confirmation, verification, and closeout records. Companion to the working paper 'J-Lens in Ouro: A Confirmed Late-Pass Band and a Target-Dependent Early-Pass Deficit'.

Training-free loop-transformer engine for GB10 — RK latent iteration for HF decoders, KV-cache-correct generation

Code and data for the paper "Stable Because It Stopped Computing"

Is the graph doing anything? A sourced survey of recurrent self-routing neural graphs as an alternative to layered networks (SR-2026-006)

Agent-first index of the looped / recurrent-depth transformer literature. The unit of knowledge is the claim instance, not the paper.