Newest papers

When Copying Is Hard: Copy-Constrained Decoding for Exact Span Reproduction

Jinghui Zhang, Lang Gao, Zongfang Liu, Ruihong Zeng, Zirui Song, Rui Yan, Kentaro Inui, Xiuying Chen · 2026 · arXiv (Cornell University)

DOI: 10.48550/arxiv.2610.28393 · cited by 0 works (OpenAlex) · open-access PDF

Indonesian

Coptain Copy adalah pengulangan bebas dari spektrum target dalam konteks yang panjang adalah tes kemampuan dasar untuk LLMs: bahkan ketika string target hadir dalam konteks, model sering masih mengalir tepat dalam spektrum atau gagal untuk berhenti dengan benar, mengungkapkan batasan dalam pengulangan sumber dan kontrol batas. Kemampuan ini masih memiliki relevansi praktis yang luas. Sebagai contoh, generasi kode sering memerlukan embedding string sensitif seperti API keywords persis seperti yang mereka muncul dalam konteks. Dalam pekerjaan ini, kami membangun benchmark untuk menguji kemampuan reproduksi spektrum tepat di antara kontrak, literatur ilmiah, kode, dan urutan acak, dan menemukan bahwa model saat ini masih membuat kesalahan yang mengejutkan dan sering. Untuk mengatasi ini, kami mengusulkan CopyGen, sebuah plugin ringan yang membatasi copy-generasi seperti untuk LLMs beku yang mengubahnya dari subproduk dari kontrol tok

Machine translation (Meta M2M100); the English original follows.

Abstract

Precisely reproducing a target span verbatim within a long context is a fundamental capability test for LLMs: even when the target string is present in context, models often still drift within the span or fail to stop correctly, revealing limitations in source grounding and boundary control. This capability has broad practical relevance. For instance, code generation often requires embedding sensitive strings such as API keys exactly as they appear in context. In this work, we construct a benchmark to test exact span reproduction ability across contracts, scientific literature, code, and random sequences, and find that current models still make surprisingly frequent errors. To address this, we propose CopyGen, a lightweight copy-constrained plugin for frozen LLMs that turns copying from a byproduct of next-token prediction into a controlled process. Our key idea is to decompose copying into three distinct decisions: when to copy, how far to copy, and when to stop. Concretely, CopyGen first determines whether to enter copy mode, then predicts a tentative copy length, and finally verifies the generated span token by token. During verification, we adaptively adjust the logits of copy continuation tokens to incentivize or suppress continuation, and emit only the longest valid prefix. Across three LLM backbones and four benchmark tasks, CopyGen improves average exact match by 35% relative without model finetuning, and integrates seamlessly with SFT to further improve it by 4%, while substantially accelerating generation. These results suggest that reliable exact copying is not something LLMs obtain for free, but a decoding-time control problem that requires explicit modeling. Code is available at https://github.com/mbzuai-nlp/CopyGen.

Read in: English · Spanish · Chinese · Hindi · Arabic · Portuguese · French · German · Japanese · Russian · Korean · Indonesian · Vietnamese

Topics: Copying, Computer science, Benchmark (surveying), Context (archaeology), Decoding methods, String (physics)

Sources: OpenAlex (CC0), arXiv, Crossref. Agent Exchange