[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-273349-105":59,"doc-detail-273349-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","scope-a-generative-approach-for-llm-prompt-compression","SCOPE - A Generative Approach for LLM Prompt Compression","","Prompt compression reduces LLM input length to control inference cost, latency, and context-limit overflow while preserving generation quality. SCOPE introduces a training-free generative framework based on chunk-level rewriting: prompts are split into semantically coherent chunks, each chunk is rewritten via summarization into a more concise form, and then reconstructed into a coherent prompt. Optimization techniques enhance critical information retention and text coherence, enabling finer control of the compression ratio. Evaluations on question-answering and summarization show consistent gains over selective compression baselines, especially at high compression ratios.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/scope-a-generative-approach-for-llm-prompt-compression/273349/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/scope-a-generative-approach-for-llm-prompt-compression/273349.png","ImageObject",300,407,{"name":92,"@type":93},"Logic","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-10-29","2026-09-15",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",17,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What problem does SCOPE address in LLM applications?","Question",{"text":113,"@type":114},"SCOPE targets the high inference cost and latency caused by long input contexts, along with the risk of exceeding model context limits.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How does SCOPE perform prompt compression?",{"text":118,"@type":114},"SCOPE splits the prompt into semantically coherent chunks, rewrites chunks using a summarization model with dynamically chosen compression ratios, and then reassembles the rewritten chunks in the original order.",{"name":120,"@type":111,"acceptedAnswer":121},"Why is SCOPE considered training-free and efficient?",{"text":122,"@type":114},"SCOPE does not require additional training; it relies on lightweight pretrained embedding and summarization models during compression, avoiding large LLM calls and keeping resource usage practical.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},273349,1789470998,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":140,"language":141,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":142,"faqs":143,"seo_title":144,"seo_description":67,"update_tm":130,"read_time":36,"preview_page_count":4},1099513958762,"https://ap-avatar.wpscdn.com/avatar/1000023916a998db790?x-image-process=image/resize,m_fixed,w_180,h_180&k=1784791008015729253","SCOPE: A Generative Approach for LLM Prompt Compression  \nTinghui Zhang  \nUniversity of Florida [tinghui.zhang@ufl.edu](tinghui.zhang@ufl.edu)  \nYifan Wang ∗  \nUniversity of Hawaii-Manoa[yifanw@hawaii.edu](yifanw@hawaii.edu)  \nDaisy Zhe Wang  \nUniversity of Florida [daisyw@cise.ufl.edu](daisyw@cise.ufl.edu)  \narXiv : 2508 . 158 13v2 [cs .CL] 20 Aug 2026  \nAbstract  \nA big issue in modern LLM applications is they tend to feed long context to LLM, which results in high inference cost and latency, and may exceed the context limit. Prompt compression addresses this issue by reducing the length of input context with minimum loss of generation quality, i.e, the goal of prompt compression is to shorten the LLM input while maintaining a high generation quality. To overcome these limitations, we propose SCOPE, a training-free generative prompt compression framework based on chunk-level rewriting. Unlike the existing token removal methods, our method centers at a chunking-and-summarization mechanism. Specifically, SCOPE splits a prompt into semantically coherent chunks and rewrites the chunks to be more concise. Then the chunks are reconstructed into a meaningful prompt. Additionally, we design several optimization techniques for SCOPE, effectively preserving critical information and text coherence in compression, as well as providing finer-grained control of the compression ratio. We conduct extensive evaluation on typical LLM applications like question-answering and summarization. Results show that SCOPE consistently outperforms the evaluated selective compression baselines across most settings, with particularly strong gains at high compression ratios.  \n1 Introduction  \nWith the increasing length of input context in Large Language Models (LLMs) applications, users/developers have to face significantly inference cost and latency increase, as well as risks exceeding model context limits. To address these issues, prompt compression methods have been developed to shorten prompt while retaining necessary information. The stateof-the-art methods are usually based on selective compression, which focus on identifying and retaining the most important tokens of the original prompt, while removing less critical tokens. Although such approaches (called “selective” or “token removal” methods in this paper) reduce prompt length, they have notable drawbacks: (1) Breaking contextual integrity: By removing tokens of the prompt, selective methods disrupt the original context, leading to a loss of coherence and a potentially lost conjunction part of different chunks.  \n(2) Limiting context scope: Selective methods use token as the processing unit, instead of considering at a higher level of the context, which limits their ability to understand the broader semantics. (3) Losing key information: Removal of tokens potentially discards key information, which never be compensated after the compression.  \nTo address the problems, we propose SCOPE, a generative prompt compression method based on prompt rewrite with generative model. 1 Specifically, SCOPE rewrites the original prompt to be more concise, while preserving the overall meaning and critical information. To balance higher-level semantics and lower-level details, the rewrite is conducted on chunklevel (i.e., chunk by chunk) rather than the whole context. SCOPE deploys a summarization model for the rewrite. And because the goal of prompt compression is to reduce LLM cost, it makes no sense to use another LLM for the summarization in our method, therefore a smaller summarization model (e.g., BART (Lewis et al. (2019))) is chosen.  \n∗ Corresponding author.  \n1 Code is available at [https://github.com/tzhan024/scope-prompt-compression](https://github.com/tzhan024/scope-prompt-compression).  \nOur method involves the following steps: (1) SCOPE first breaks down the original prompt into semantically coherent chunks (size not fixed), ensuring each chunk covers a complete and meaningful piece of information.","cbCaibx4gYacY0q3","https://ap.wps.com/l/cbCaibx4gYacY0q3","pdf",321241,16,"English","# Abstract\n# 1 Introduction\n## Problem of long context and selective compression drawbacks\n## SCOPE framework overview and chunk-level process\n## Comparison with related work and efficiency advantages","[{\"question\":\"What problem does SCOPE address in LLM applications?\",\"answer\":\"SCOPE targets the high inference cost and latency caused by long input contexts, along with the risk of exceeding model context limits.\"},{\"question\":\"How does SCOPE perform prompt compression?\",\"answer\":\"SCOPE splits the prompt into semantically coherent chunks, rewrites chunks using a summarization model with dynamically chosen compression ratios, and then reassembles the rewritten chunks in the original order.\"},{\"question\":\"Why is SCOPE considered training-free and efficient?\",\"answer\":\"SCOPE does not require additional training; it relies on lightweight pretrained embedding and summarization models during compression, avoiding large LLM calls and keeping resource usage practical.\"}]","SCOPE - A Generative Approach for LLM Prompt Compression | PDF"]