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.

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.

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

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

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)

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

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

Recurrent-depth, P/R/C looped transformer with MoE, GQA, and stable recurrence

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