"Skip to main content\narXiv is now an independent nonprofit!\nLearn more\n×\nSearch\nSubmit\nDonate\nLog in\nShowing 1–1 of 1 results for all: attention sink KV cache token eviction long context serving\nSearch v0.5.6 released 2020-02-24 \nSearch term or terms\nAll fields\nTitle\nAuthor(s)\nAbstract\nComments\nJournal reference\nACM classification\nMSC classification\nReport number\narXiv identifier\nDOI\nORCID\nLicense (URI)\narXiv author ID\nHelp pages\nFull text\nSearch\n Show abstracts Hide abstracts\nAdvanced Search\n25\n50\n100\n200\n results per page.\nSort results by \nAnnouncement date (newest first)\nAnnouncement date (oldest first)\nSubmission date (newest first)\nSubmission date (oldest first)\nRelevance\nGo\n\narXiv:2512.03324 [pdf, ps, other] \n\n cs.LG cs.AI\n\nCache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs\n\nAuthors: Ngoc Bui, Shubham Sharma, Simran Lamba, Saumitra Mishra, Rex Ying\n\nAbstract: Memory and computation remain core bottlenecks in long-horizon LLM inference due to the quadratic cost of self-… ▽ More\n\nSubmitted 28 February, 2026; v1 submitted 2 December, 2025; originally announced December 2025.\n\nWe gratefully acknowledge support from our major funders, member institutions, and all contributors.\nAbout\n·\nHelp\n·\nContact\n·\nSubscribe\n·\nCopyright\n·\nPrivacy\n·\nAccessibility\n·\nOperational Status\n(opens in new tab)\nMajor funding support from"