` 包裹，长文用 `
`/`
` 划分小节。 * 规则3： 列表内容用 `
`/`
- `，引用用 `
`，代码用 `
`/``，重要句子可用 ``。
* 规则4： 保留原文所有信息，不得增删改内容。
* 规则5： 不要输出 ``/``/`` 等文档级标签。
* 输入文本： 一段关于LLM智能体、KV缓存淘汰、PACE算法的科学新闻/技术文章。
2. 分析输入文本结构及内容：
* *第1句（问题引入）：* "Long-horizon LLM agents accumulate sessions of up to millions of tokens, far more key-value (KV) cache than one GPU holds, so serving systems keep a small budget of tokens and evict the rest." -> 这是一个独立段落。
* *第2句（现有方法及目标）：* "Eviction methods differ in which tokens they keep but share one goal: that the tokens a later step needs are still cached." -> 可以和第1句合并，或者作为独立段落。让我们先看一下后面。其实把它作为引入段落的一部分，或者用 `` 引入一个新小节都不错。
* *第3句（研究发现/问题揭示）：* "We show this is not enough. In 100K-token coding sessions on Qwen3-8B with a 2,048-token budget, a planner-guided policy that keeps over 99% of the needed tokens resident still fails late steps increasingly often: its accuracy falls from 93% to 64% as the gap between a definition and its use grows from 2K to 90K tokens, at residency 0.996. The needed tokens are resident but unretrievable: what grows is their rotary-position distance to the query." -> 这是一个巨大段落。分为：
* "We show this is not enough."
* "In 100K-token coding sessions..."
* "The needed tokens are resident but unretrievable..."
* *第4句（提出解决方案）：* "We present PACE (Planner-Aware Context Engine), a KV-cache engine that couples plan-aware retention, in which the agent's planner tells the cache which context future steps will read, with bounded positions, which renumber kept tokens so their rotary distance to the query stays below 8K at any session length." -> 提出 PACE 的核心思想。
* *第5句（初步评估结果）：* "In a pre-registered evaluation on fresh tasks, PACE passes 97% of hidden unit tests at 100K tokens, against 42% for H2O and 3% for SnapKV under the same budget, and needs both components: either alone reaches 63% (the combination beats each, p < 10^-4)." -> 实验结果。
* *第6句（扩展性/长上下文结果）：* "The result holds with length: in 1M- and 2M-token sessions on one 24 GB GPU, with definitions written 0.9M and 1.8M tokens before use, PACE passes 99% and 100% of checks and recalls every session fact, while bounded positions without foresight falls from 33% to 13%; PACE holds under 0.7 GB of KV state where the full cache would need 147-295 GB." -> 长上下文表现与内存占用。
* *第7句（性能开销）：* "Renumbering is exact, costs no measurable perplexity, and adds 9% wall time." -> 性能开销。
* *第8句（其他模型结果）：* "On Mistral-7B, whose coding ability caps it at 40% even with nothing evicted, PACE stays at that ceiling while retention alone falls below it." -> Mistral-7B 上的结果。
3. 拟定排版结构：
* 由于文本不长，可以分为 3-4 个小节。
* *小节 1：背景与问题*
* `
` 背景与问题背景 (Background & The Problem) —— 可以用英文标题，配合内容的专业性。就用英文 `
Background and the Problem
`。
* 第1句和第2句作为 ``。
* 第3句拆分一下，"We show this is not enough." 可以加粗。
* *小节 2：PACE 系统*
* `
The PACE Solution
`
* 第4句介绍 PACE。
* *小节 3：评估与性能*
* `Evaluation and Performance
`
* 第5、6、7、8句。第5、6句各起一段。第7、8句可以合并或分开。分开会清晰一点。
4. 调整结构以严格遵守规则：
* Rule 2: "每个自然段落用包裹，长文用
/
划分小节" (Wrap each natural paragraph with `
`, use `
`，代码用 ``/``，重要句子可用 ``。 * 规则4： 保留原文所有信息，不得增删改内容。 * 规则5： 不要输出 ``/``/`` 等文档级标签。 * 输入文本： 一段关于LLM智能体、KV缓存淘汰、PACE算法的科学新闻/技术文章。 2. 分析输入文本结构及内容： * *第1句（问题引入）：* "Long-horizon LLM agents accumulate sessions of up to millions of tokens, far more key-value (KV) cache than one GPU holds, so serving systems keep a small budget of tokens and evict the rest." -> 这是一个独立段落。 * *第2句（现有方法及目标）：* "Eviction methods differ in which tokens they keep but share one goal: that the tokens a later step needs are still cached." -> 可以和第1句合并，或者作为独立段落。让我们先看一下后面。其实把它作为引入段落的一部分，或者用 `` 引入一个新小节都不错。 * *第3句（研究发现/问题揭示）：* "We show this is not enough. In 100K-token coding sessions on Qwen3-8B with a 2,048-token budget, a planner-guided policy that keeps over 99% of the needed tokens resident still fails late steps increasingly often: its accuracy falls from 93% to 64% as the gap between a definition and its use grows from 2K to 90K tokens, at residency 0.996. The needed tokens are resident but unretrievable: what grows is their rotary-position distance to the query." -> 这是一个巨大段落。分为： * "We show this is not enough." * "In 100K-token coding sessions..." * "The needed tokens are resident but unretrievable..." * *第4句（提出解决方案）：* "We present PACE (Planner-Aware Context Engine), a KV-cache engine that couples plan-aware retention, in which the agent's planner tells the cache which context future steps will read, with bounded positions, which renumber kept tokens so their rotary distance to the query stays below 8K at any session length." -> 提出 PACE 的核心思想。 * *第5句（初步评估结果）：* "In a pre-registered evaluation on fresh tasks, PACE passes 97% of hidden unit tests at 100K tokens, against 42% for H2O and 3% for SnapKV under the same budget, and needs both components: either alone reaches 63% (the combination beats each, p < 10^-4)." -> 实验结果。 * *第6句（扩展性/长上下文结果）：* "The result holds with length: in 1M- and 2M-token sessions on one 24 GB GPU, with definitions written 0.9M and 1.8M tokens before use, PACE passes 99% and 100% of checks and recalls every session fact, while bounded positions without foresight falls from 33% to 13%; PACE holds under 0.7 GB of KV state where the full cache would need 147-295 GB." -> 长上下文表现与内存占用。 * *第7句（性能开销）：* "Renumbering is exact, costs no measurable perplexity, and adds 9% wall time." -> 性能开销。 * *第8句（其他模型结果）：* "On Mistral-7B, whose coding ability caps it at 40% even with nothing evicted, PACE stays at that ceiling while retention alone falls below it." -> Mistral-7B 上的结果。 3. 拟定排版结构： * 由于文本不长，可以分为 3-4 个小节。 * *小节 1：背景与问题* * `
` 背景与问题背景 (Background & The Problem) —— 可以用英文标题，配合内容的专业性。就用英文 `
Background and the Problem
`。 * 第1句和第2句作为 ``。 * 第3句拆分一下，"We show this is not enough." 可以加粗。 * *小节 2：PACE 系统* * `
The PACE Solution
` * 第4句介绍 PACE。 * *小节 3：评估与性能* * `Evaluation and Performance
` * 第5、6、7、8句。第5、6句各起一段。第7、8句可以合并或分开。分开会清晰一点。 4. 调整结构以严格遵守规则： * Rule 2: "每个自然段落用包裹，长文用
/
划分小节" (Wrap each natural paragraph with `
`, use `
`/`
` to divide sections in long texts). * 不要打乱原文的句子顺序，除非为了排版加个标题。既然是单篇报道，加几个 `
` 标签划分非常合适。 让我们精细化HTML结构： `
Background and the Problem
` `Long-horizon LLM agents accumulate sessions of up to millions of tokens, far more key-value (KV) cache than one GPU holds, so serving systems keep a small budget of tokens and evict the rest. Eviction methods differ in which tokens they keep but share one goal: that the tokens a later step needs are still cached.
` `We show this is not enough. In 100K-token coding sessions on Qwen3-8B with a 2,048-token budget, a planner-guided policy that keeps over 99% of the needed tokens resident still fails late steps increasingly often: its accuracy falls from 93% to 64% as the gap between a definition and its use grows from 2K to 90K tokens, at residency 0.996. The needed tokens are resident but unretrievable: what grows is their rotary-position distance to the query.
` `The PACE System
` `We present PACE (Planner-Aware Context Engine), a KV-cache engine that couples plan-aware retention, in which the agent's planner tells the cache which context future steps will read, with bounded positions, which renumber kept tokens so their rotary distance to the query stays below 8K at any session length.
` `Evaluation Results
` `In a pre-registered evaluation on fresh tasks, PACE passes 97% of hidden unit tests at 100K tokens, against 42% for H2O and 3% for SnapKV under the same budget, and needs both components: either alone reaches 63% (the combination beats each, p < 10^-4).
` `The result holds with length: in 1M- and 2M-token sessions on one 24 GB GPU, with definitions written 0.9M and 1.8M tokens before use, PACE passes 99% and 100% of checks and recalls every session fact, while bounded positions without foresight falls from 33% to 13%; PACE holds under 0.7 GB of KV state where the full cache would need 147-295 GB.
` `Renumbering is exact, costs no measur