https://blog.looprouter.com/tags/%E9%95%BF%E4%B8%8A%E4%B8%8B%E6%96%87/Recent content in 长上下文 on LoopRouter BlogHugozh-cnSun, 20 Sep 2026 00:00:00 +0000https://blog.looprouter.com/docs/on-demand-attention-long-context-inference-cost/Sun, 20 Sep 2026 00:00:00 +0000https://blog.looprouter.com/docs/on-demand-attention-long-context-inference-cost/<blockquote> <p>本文基于 arXiv 论文《On-Demand Attention: Language Models Know When to Recall》(2026-09-17 提交)的公开摘要整理与二次分析,加入 LoopRouter 的接入/成本视角;原文见文末链接。文中不含摘要之外的量化数字,具体收益请以完整论文与自有负载复测为准。</p>https://blog.looprouter.com/docs/deepseek-v4-1-flash-kv-cache-cost/Fri, 11 Sep 2026 00:00:00 +0000https://blog.looprouter.com/docs/deepseek-v4-1-flash-kv-cache-cost/<blockquote> <p>本文基于 InfoQ 中文(2026-09-10,作者褚杏娟)对 DeepSeek 技术报告的报道整理与二次分析,加入 LoopRouter 的模型选型与成本视角;数据均转述自该报道,笔者未独立复现原始报告。原文见文末链接。</p>