ResidualQuant introduces a 2‑bit residual quantization scheme for KV caches in looped Transformers, leveraging the similarity of KV states across loops. By using the final‑loop KV states as a reference and applying least‑square scaling, rotations, and loop‑wise mixed precision, the method achieves accurate quantization down to INT2. Experiments on various looped Transformer models show that ResidualQuant improves the accuracy‑memory tradeoff, retaining near BF16 accuracy while reducing KV storage by 80.7% and boosting decode throughput and batch size on an RTX 5090.
By Heejun Kim, Junyoung Lee, SangLyul Cho, Dongsu Han, Insu Han, Sehoon Kim
arXiv:2610.16368v1 Announce Type: cross
Abstract: Looped Language Models (LoopLMs) improve parameter efficiency by recursively reusing shared Transformer modules, yet their repeated computation over...
By Zhixiong Zhao, Zukang Xu, Xing Hu, Lifeng Liu, Guangyu Sun, Dawei Yang
Looped transformers are a parameter-efficient alternative to standard transformers for language modeling. By reapplying the same layers across multiple loop steps, recent looped language models match...
arXiv:2608. 08888v1 Announce Type: new Abstract: Autoregressive transformers compute along two axes: horizontally across generated tokens, and vertically through model depth.
By Xi Wang, Ziyang Cai, Zheng Zhan, Harry Dong, Ying Fan, Gustavo de Rosa, Tim Pearce, John Langford
The paper investigates how looping—reusing transformer weights to deepen models without adding parameters—affects performance under different constraints. Experiments on dense decoder‑only transformers up to 1 B parameters show that looping the full transformer yields best quality for a fixed parameter budget, while looping a middle block is preferable when training FLOPs are limited. Adding loop‑specific low‑rank adapters improves quality, but decoding slows linearly with loops; a parallel loop design that passes hidden states across token positions can mitigate this slowdown, though careful handling of token embeddings is needed to retain gains.
By Chanakya Ekbote, Artidoro Pagnoni, Devendra Singh Sachan, Shruti Bhosale, Mike Lewis, Maryam Fazel-Zarandi, Shashank Rajput, Sujan Kumar Gonugondla
The paper introduces TRACE, a framework that embeds a Vector Quantizer at the post‑attention residual of each Transformer layer, enabling exchanged activations to be encoded as compact codebook indices. During inference, receivers recover key/value representations via table lookup, eliminating downstream computation and allowing communication–computation overlap. A hierarchical codebook further permits runtime bitrate adaptation, and experiments on ViT‑Base and ViViT‑B demonstrate up to 1.89% accuracy gain, 1.27× speedup, and a 22,624× reduction in communication overhead compared to uncompressed inference.
By Muhammad Azlan Qazi, Alexandros Iosifidis, Qi Zhang
arXiv:2602.08005v2 Announce Type: replace-cross
Abstract: Efficient long-context inference faces two coupled bottlenecks: KV-cache memory grows linearly with context length, while attention computati...
By Jitai Hao, Qiang Huang, Yaowei Wang, Min Zhang, Jun Yu
The paper "LLM Inference in a Flash!" proposes an integer‑only quantization scheme and a dictionary‑based KV cache compression technique to enable large language model inference on compute‑in‑flash (CIF) devices. By eliminating floating‑point operations and reducing KV cache traffic through sparse dictionary coding, the authors achieve minimal accuracy loss while cutting dynamic KV cache traffic by 15× on Llama‑3.1‑8B and Qwen‑2.5‑7B models.
By Sebastian Zhao, Minseo Kim, Coleman Hooper, Luca Manolache, Michael W. Mahoney, Yakun Sophia Shao, Kurt Keutzer, Amir Gholami
The paper introduces LoopCD, a training‑free contrastive decoding framework that improves token selection in Loop‑Transformer models by comparing the final prediction with earlier recurrent passes. LoopCD operates either in logit space (LoopCD‑Logits) with a single extra output pass or in hidden‑state space (LoopCD‑Hidden) with no output overhead. Across multiple looped Transformer families, LoopCD yields significant performance gains—raising pass@1 scores on tasks such as AIME 2024 and HumanEval—while enabling a reduction in the number of recurrent loops and a corresponding decrease in inference FLOPs.
By Weihao Liu, Huangjie Zheng, Tianrong Chen, Rohit Dilip, Richard He Bai, Yizhu Jiao, Yuyang Wang, Ruixiang Zhang
The paper introduces an LSTM-style transformer that shares a single key–value cache across its recurrent core, reducing parameters by 40.4% compared to an untied model while keeping forward FLOPs similar. It aligns parallel and continuation training passes with a consistency loss, eliminating 81–89% of the continuation‑loss penalty on FineWeb. Experiments across multiple widths and token budgets show that weight tying incurs the smallest loss cost at lower token budgets, highlighting its potential for data‑ and storage‑constrained settings.
By Andreas Kirsch
arXiv:2607. 27600v1 Announce Type: new Abstract: Key-value (KV) cache management through compression and eviction strategies has emerged as an important research direction in recent years.
By Stephen Gould, Anton van den Hengel
arXiv:2607. 01831v1 Announce Type: cross Abstract: Long-context inference is increasingly common in large language model (LLM) serving, driven by retrieval-augmented generation and agentic systems.
By Wenchen Han, Gingfung Matthew Yeung, Marco Barletta, William Toner, Amory Hoste, Adam Barker