The paper examines whether context compression, which encodes inputs into compact representations while preserving semantic content, maintains refusal alignment in instruction-tuned language models. It finds that encoder-based compression consistently weakens refusal behavior across various model families and architectures, even though reconstruction fidelity remains high. Mechanistic analysis attributes this to attenuation of activation along the decoder’s learned refusal direction, and introduces Memory Steering—a lightweight inference-time intervention that restores refusal rates without retraining.
By Anthony Nguyen, Wenjun Lin
The paper reports a Size‑Fidelity Paradox in large‑language‑model (LLM) context compression: larger compressors can produce lower reconstruction error yet yield less faithful reconstructed contexts. Two key mechanisms drive this effect: knowledge overwriting, where larger models replace source facts with their own beliefs, and semantic drift, where they paraphrase or restructure content. Across 27 compressor setups, mid‑sized models often recover contexts more faithfully, and analysis of embedding geometry shows that larger models organize memory into broader semantic subspaces, leading to ambiguous representations that weaken recovery.
By Ruishan Guo, Yibing Liu, Guoxin Ma, Yan Wang, Yueyang Zhang, Long Xia, Kecheng Chen, Zhiyuan Sun, Daiting Shi
Modern large language models are pretrained on massive datasets, making it difficult to prevent benchmark data from entering their training sets and undermining the reliability of evaluation results....
arXiv:2609.27510v1 Announce Type: cross
Abstract: Modern large language models are pretrained on massive datasets, making it difficult to prevent benchmark data from entering their training sets and...
By Kaifeng Tan, Yudong Li, Linlin Shen
arXiv:2609.25537v1 Announce Type: new
Abstract: Large language model (LLM) inference is constrained by the quadratic scaling of self-attention and the linear scaling of the KV cache, increasing laten...
By Md Mostafizer Rahman, Md Faizul Ibne Amin, Md Shahajada Mia, Yutaka Watanobe, Fang Liu
Long-context language model inference is bottlenecked by memory, as the KV cache grows with context length. Recent techniques to compress the KV cache fall short: they either degrade model quality substantially or require considerable time and compute to compress a single long prompt.
arXiv:2606. 09659v1 Announce Type: cross Abstract: Long-context language model inference is bottlenecked by memory, as the KV cache grows with context length.
By Ang Li, Sean McLeish, Haozhe Chen, Nimit Kalra, Zaiqian Chen, Artem Gazizov, Venkata Anoop Suhas Kumar Morisetty, Bhavya Kailkhura, Harshitha Menon, Zhuang Liu, Brian R. Bartoldson, Tom Goldstein, Sanae Lotfi, Micah Goldblum, Pavel Izmailov
arXiv:2608. 11249v1 Announce Type: cross Abstract: We study the problem of lossless text compression, motivated by the rapid growth in the collection and storage of digital textual data - including plain text, source code, and structured formats such as XML - and by recent advances in neural language model-based compression.
By Angelo Nardone, Paolo Ferragina
The paper introduces Variance‑Calibrated Modulation (VCM), a training‑free pre‑decoding technique that reshapes language model probability distributions before truncation. VCM uses two dynamic mechanisms: a Contextual Searchlight via PMI to suppress stopwords and highlight context‑relevant tokens, and an Adaptive Self‑Debiasing that applies scale‑invariant penalization based on real‑time logit standard deviation. Experiments on open‑ended generation, factual QA, and mathematical reasoning show that VCM consistently reduces the likelihood trap, improving diversity, coherence, and reasoning accuracy with minimal computational cost.
By Yuanhao Ding, Meimingwei Li, Esteban Garces Arias, Matthias A{\ss}enmacher, Christian Heumann, Chongsheng Zhang
arXiv:2608.00311v2 Announce Type: replace
Abstract: Long-context inference with large language models (LLMs) is costly: self-attention during prefill scales quadratically with sequence length, and th...
By Maryam Haghifam, Jason Cong, Yizhou Sun
arXiv:2610.10304v1 Announce Type: new
Abstract: We study whether latent sequence compression of prompt prefixes preserves the capabilities that large language models rely on during inference. We intr...
By Mingyan Liu, Min Huang
The paper introduces zip2zip++, a method for inference‑time token compression that reduces token counts using a dynamic Lempel‑Ziv‑Welch (LZW) codec while preserving language model quality. It identifies four problems with prior LZW‑based compression—embedding conflict, improper RoPE application, harmful digit merges, and segmentation ambiguity—and proposes solutions: separate input/output embeddings, size‑aware RoPE, and a selective‑merge LZW codec. Experiments on four models show that these changes halve the quality gap in perplexity and reasoning accuracy compared to the uncompressed baseline, achieving similar compression ratios.
By Saibo Geng, Andrea Gentilini, Xinxian Ma, Nathan Ranchin, Yunzhen Yao, Robert West