IAAILoopUS
LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models
Looped Depth Up-Scaling converts a pretrained LLM into an encoder, a looped reasoning block and a decoder, using representation-guided block decomposition, an input-dependent selective gate against drift, random deep supervision, and a confidence head for early exit.
Inside the method
Encoder layers → [ Looped reasoning block + selective gate ] × r (confidence exit) → Decoder layers
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Core recurrence
- Depth control
- Adaptive early exit via confidence head
- KV / state strategy
- See paper
Reading note
Improves reasoning-oriented performance without longer generated traces or recurrent pretraining from scratch; released conversions include Qwen3 and Phi-4 backbones.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{loopus2026,
title = {LoopUS: Recasting Pretrained LLMs into Looped Latent Refinement Models},
author = {Taekhyun Park and Yongjae Lee and Dohee Kim and Hyerim Bae},
year = {2026},
eprint = {2605.11011},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2605.11011}
}