IAAILoopQuant
LoopQuant: Accurate and Efficient Post-Training Quantization for Looped Transformers
A loop-aware post-training quantization framework for looped language models. It canonicalizes activation distributions across loops while keeping weights shared, allocates activation precision per loop under a fixed bit budget, and calibrates the quantized model along its own low-bit recursive trajectory.
Inside the method
[ Shared block ] × R with per-loop activation precision
Simplified conceptual schematic. Consult the paper for the complete architecture.
- Recurrence family
- Full-stack recurrence
- Depth control
- Fixed loops of the base model
- KV / state strategy
- See paper
Reading note
On Ouro-2.6B, W4A4 LoopQuant beats all W4A8 baselines, cuts model memory by 65.3% with 3.4% extra storage over naive INT4, and decodes 1.7× faster than BF16.
Sources checked 2026-10-28. This catalog does not imply independent reproduction.
Related works
Cite this work
@misc{loopquant2026,
title = {LoopQuant: Accurate and Efficient Post-Training Quantization for Looped Transformers},
author = {Zhixiong Zhao and Zukang Xu and Xing Hu and Lifeng Liu and Guangyu Sun and Dawei Yang},
year = {2026},
eprint = {2610.16368},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2610.16368}
}