mirror of
https://github.com/headroomlabs-ai/headroom.git
synced 2026-08-27 14:17:10 -04:00
3571 lines
No EOL
129 KiB
HTML
3571 lines
No EOL
129 KiB
HTML
|
|
<!doctype html>
|
|
<html lang="en" class="no-js">
|
|
<head>
|
|
|
|
<meta charset="utf-8">
|
|
<meta name="viewport" content="width=device-width,initial-scale=1">
|
|
|
|
<meta name="description" content="The Context Optimization Layer for LLM Applications">
|
|
|
|
|
|
|
|
<link rel="canonical" href="https://chopratejas.github.io/headroom/transforms/">
|
|
|
|
|
|
<link rel="prev" href="../text-compression/">
|
|
|
|
|
|
<link rel="next" href="../ARCHITECTURE/">
|
|
|
|
|
|
|
|
|
|
|
|
<link rel="icon" href="../assets/images/favicon.png">
|
|
<meta name="generator" content="mkdocs-1.6.1, mkdocs-material-9.7.1">
|
|
|
|
|
|
|
|
<title>Transforms - Headroom</title>
|
|
|
|
|
|
|
|
<link rel="stylesheet" href="../assets/stylesheets/main.484c7ddc.min.css">
|
|
|
|
|
|
<link rel="stylesheet" href="../assets/stylesheets/palette.ab4e12ef.min.css">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin>
|
|
<link rel="stylesheet" href="https://fonts.googleapis.com/css?family=Roboto:300,300i,400,400i,700,700i%7CRoboto+Mono:400,400i,700,700i&display=fallback">
|
|
<style>:root{--md-text-font:"Roboto";--md-code-font:"Roboto Mono"}</style>
|
|
|
|
|
|
|
|
<script>__md_scope=new URL("..",location),__md_hash=e=>[...e].reduce(((e,_)=>(e<<5)-e+_.charCodeAt(0)),0),__md_get=(e,_=localStorage,t=__md_scope)=>JSON.parse(_.getItem(t.pathname+"."+e)),__md_set=(e,_,t=localStorage,a=__md_scope)=>{try{t.setItem(a.pathname+"."+e,JSON.stringify(_))}catch(e){}}</script>
|
|
|
|
|
|
|
|
|
|
|
|
</head>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<body dir="ltr" data-md-color-scheme="default" data-md-color-primary="indigo" data-md-color-accent="indigo">
|
|
|
|
|
|
<input class="md-toggle" data-md-toggle="drawer" type="checkbox" id="__drawer" autocomplete="off">
|
|
<input class="md-toggle" data-md-toggle="search" type="checkbox" id="__search" autocomplete="off">
|
|
<label class="md-overlay" for="__drawer"></label>
|
|
<div data-md-component="skip">
|
|
|
|
|
|
<a href="#transform-reference" class="md-skip">
|
|
Skip to content
|
|
</a>
|
|
|
|
</div>
|
|
<div data-md-component="announce">
|
|
|
|
</div>
|
|
|
|
|
|
|
|
|
|
<header class="md-header" data-md-component="header">
|
|
<nav class="md-header__inner md-grid" aria-label="Header">
|
|
<a href=".." title="Headroom" class="md-header__button md-logo" aria-label="Headroom" data-md-component="logo">
|
|
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 8a3 3 0 0 0 3-3 3 3 0 0 0-3-3 3 3 0 0 0-3 3 3 3 0 0 0 3 3m0 3.54C9.64 9.35 6.5 8 3 8v11c3.5 0 6.64 1.35 9 3.54 2.36-2.19 5.5-3.54 9-3.54V8c-3.5 0-6.64 1.35-9 3.54"/></svg>
|
|
|
|
</a>
|
|
<label class="md-header__button md-icon" for="__drawer">
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M3 6h18v2H3zm0 5h18v2H3zm0 5h18v2H3z"/></svg>
|
|
</label>
|
|
<div class="md-header__title" data-md-component="header-title">
|
|
<div class="md-header__ellipsis">
|
|
<div class="md-header__topic">
|
|
<span class="md-ellipsis">
|
|
Headroom
|
|
</span>
|
|
</div>
|
|
<div class="md-header__topic" data-md-component="header-topic">
|
|
<span class="md-ellipsis">
|
|
|
|
Transforms
|
|
|
|
</span>
|
|
</div>
|
|
</div>
|
|
</div>
|
|
|
|
|
|
<form class="md-header__option" data-md-component="palette">
|
|
|
|
|
|
|
|
|
|
<input class="md-option" data-md-color-media="" data-md-color-scheme="default" data-md-color-primary="indigo" data-md-color-accent="indigo" aria-label="Switch to dark mode" type="radio" name="__palette" id="__palette_0">
|
|
|
|
<label class="md-header__button md-icon" title="Switch to dark mode" for="__palette_1" hidden>
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 8a4 4 0 0 0-4 4 4 4 0 0 0 4 4 4 4 0 0 0 4-4 4 4 0 0 0-4-4m0 10a6 6 0 0 1-6-6 6 6 0 0 1 6-6 6 6 0 0 1 6 6 6 6 0 0 1-6 6m8-9.31V4h-4.69L12 .69 8.69 4H4v4.69L.69 12 4 15.31V20h4.69L12 23.31 15.31 20H20v-4.69L23.31 12z"/></svg>
|
|
</label>
|
|
|
|
|
|
|
|
|
|
|
|
<input class="md-option" data-md-color-media="" data-md-color-scheme="slate" data-md-color-primary="indigo" data-md-color-accent="indigo" aria-label="Switch to light mode" type="radio" name="__palette" id="__palette_1">
|
|
|
|
<label class="md-header__button md-icon" title="Switch to light mode" for="__palette_0" hidden>
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 18c-.89 0-1.74-.2-2.5-.55C11.56 16.5 13 14.42 13 12s-1.44-4.5-3.5-5.45C10.26 6.2 11.11 6 12 6a6 6 0 0 1 6 6 6 6 0 0 1-6 6m8-9.31V4h-4.69L12 .69 8.69 4H4v4.69L.69 12 4 15.31V20h4.69L12 23.31 15.31 20H20v-4.69L23.31 12z"/></svg>
|
|
</label>
|
|
|
|
|
|
</form>
|
|
|
|
|
|
|
|
<script>var palette=__md_get("__palette");if(palette&&palette.color){if("(prefers-color-scheme)"===palette.color.media){var media=matchMedia("(prefers-color-scheme: light)"),input=document.querySelector(media.matches?"[data-md-color-media='(prefers-color-scheme: light)']":"[data-md-color-media='(prefers-color-scheme: dark)']");palette.color.media=input.getAttribute("data-md-color-media"),palette.color.scheme=input.getAttribute("data-md-color-scheme"),palette.color.primary=input.getAttribute("data-md-color-primary"),palette.color.accent=input.getAttribute("data-md-color-accent")}for(var[key,value]of Object.entries(palette.color))document.body.setAttribute("data-md-color-"+key,value)}</script>
|
|
|
|
|
|
|
|
|
|
|
|
<label class="md-header__button md-icon" for="__search">
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M9.5 3A6.5 6.5 0 0 1 16 9.5c0 1.61-.59 3.09-1.56 4.23l.27.27h.79l5 5-1.5 1.5-5-5v-.79l-.27-.27A6.52 6.52 0 0 1 9.5 16 6.5 6.5 0 0 1 3 9.5 6.5 6.5 0 0 1 9.5 3m0 2C7 5 5 7 5 9.5S7 14 9.5 14 14 12 14 9.5 12 5 9.5 5"/></svg>
|
|
</label>
|
|
<div class="md-search" data-md-component="search" role="dialog">
|
|
<label class="md-search__overlay" for="__search"></label>
|
|
<div class="md-search__inner" role="search">
|
|
<form class="md-search__form" name="search">
|
|
<input type="text" class="md-search__input" name="query" aria-label="Search" placeholder="Search" autocapitalize="off" autocorrect="off" autocomplete="off" spellcheck="false" data-md-component="search-query" required>
|
|
<label class="md-search__icon md-icon" for="__search">
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M9.5 3A6.5 6.5 0 0 1 16 9.5c0 1.61-.59 3.09-1.56 4.23l.27.27h.79l5 5-1.5 1.5-5-5v-.79l-.27-.27A6.52 6.52 0 0 1 9.5 16 6.5 6.5 0 0 1 3 9.5 6.5 6.5 0 0 1 9.5 3m0 2C7 5 5 7 5 9.5S7 14 9.5 14 14 12 14 9.5 12 5 9.5 5"/></svg>
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M20 11v2H8l5.5 5.5-1.42 1.42L4.16 12l7.92-7.92L13.5 5.5 8 11z"/></svg>
|
|
</label>
|
|
<nav class="md-search__options" aria-label="Search">
|
|
|
|
<button type="reset" class="md-search__icon md-icon" title="Clear" aria-label="Clear" tabindex="-1">
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M19 6.41 17.59 5 12 10.59 6.41 5 5 6.41 10.59 12 5 17.59 6.41 19 12 13.41 17.59 19 19 17.59 13.41 12z"/></svg>
|
|
</button>
|
|
</nav>
|
|
|
|
<div class="md-search__suggest" data-md-component="search-suggest"></div>
|
|
|
|
</form>
|
|
<div class="md-search__output">
|
|
<div class="md-search__scrollwrap" tabindex="0" data-md-scrollfix>
|
|
<div class="md-search-result" data-md-component="search-result">
|
|
<div class="md-search-result__meta">
|
|
Initializing search
|
|
</div>
|
|
<ol class="md-search-result__list" role="presentation"></ol>
|
|
</div>
|
|
</div>
|
|
</div>
|
|
</div>
|
|
</div>
|
|
|
|
|
|
|
|
<div class="md-header__source">
|
|
<a href="https://github.com/chopratejas/headroom" title="Go to repository" class="md-source" data-md-component="source">
|
|
<div class="md-source__icon md-icon">
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512"><!--! Font Awesome Free 7.1.0 by @fontawesome - https://fontawesome.com License - https://fontawesome.com/license/free (Icons: CC BY 4.0, Fonts: SIL OFL 1.1, Code: MIT License) Copyright 2025 Fonticons, Inc.--><path d="M173.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6m-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3m44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9M252.8 8C114.1 8 8 113.3 8 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C436.2 457.8 504 362.9 504 252 504 113.3 391.5 8 252.8 8M105.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1m-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7m32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1m-11.4-14.7c-1.6 1-1.6 3.6 0 5.9s4.3 3.3 5.6 2.3c1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2"/></svg>
|
|
</div>
|
|
<div class="md-source__repository">
|
|
chopratejas/headroom
|
|
</div>
|
|
</a>
|
|
</div>
|
|
|
|
</nav>
|
|
|
|
</header>
|
|
|
|
<div class="md-container" data-md-component="container">
|
|
|
|
|
|
|
|
|
|
|
|
<nav class="md-tabs" aria-label="Tabs" data-md-component="tabs">
|
|
<div class="md-grid">
|
|
<ul class="md-tabs__list">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item">
|
|
<a href=".." class="md-tabs__link">
|
|
|
|
|
|
|
|
|
|
|
|
Home
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item">
|
|
<a href="../quickstart/" class="md-tabs__link">
|
|
|
|
|
|
|
|
Getting Started
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item">
|
|
<a href="../proxy/" class="md-tabs__link">
|
|
|
|
|
|
|
|
User Guide
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item">
|
|
<a href="../langchain/" class="md-tabs__link">
|
|
|
|
|
|
|
|
Integrations
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item md-tabs__item--active">
|
|
<a href="../llmlingua/" class="md-tabs__link">
|
|
|
|
|
|
|
|
Advanced
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item">
|
|
<a href="../api/" class="md-tabs__link">
|
|
|
|
|
|
|
|
Reference
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item">
|
|
<a href="../macos-deployment/" class="md-tabs__link">
|
|
|
|
|
|
|
|
Deployment
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-tabs__item">
|
|
<a href="../memory/" class="md-tabs__link">
|
|
|
|
|
|
|
|
Memory
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</div>
|
|
</nav>
|
|
|
|
|
|
|
|
<main class="md-main" data-md-component="main">
|
|
<div class="md-main__inner md-grid">
|
|
|
|
|
|
|
|
<div class="md-sidebar md-sidebar--primary" data-md-component="sidebar" data-md-type="navigation" >
|
|
<div class="md-sidebar__scrollwrap">
|
|
<div class="md-sidebar__inner">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<nav class="md-nav md-nav--primary md-nav--lifted" aria-label="Navigation" data-md-level="0">
|
|
<label class="md-nav__title" for="__drawer">
|
|
<a href=".." title="Headroom" class="md-nav__button md-logo" aria-label="Headroom" data-md-component="logo">
|
|
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M12 8a3 3 0 0 0 3-3 3 3 0 0 0-3-3 3 3 0 0 0-3 3 3 3 0 0 0 3 3m0 3.54C9.64 9.35 6.5 8 3 8v11c3.5 0 6.64 1.35 9 3.54 2.36-2.19 5.5-3.54 9-3.54V8c-3.5 0-6.64 1.35-9 3.54"/></svg>
|
|
|
|
</a>
|
|
Headroom
|
|
</label>
|
|
|
|
<div class="md-nav__source">
|
|
<a href="https://github.com/chopratejas/headroom" title="Go to repository" class="md-source" data-md-component="source">
|
|
<div class="md-source__icon md-icon">
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512"><!--! Font Awesome Free 7.1.0 by @fontawesome - https://fontawesome.com License - https://fontawesome.com/license/free (Icons: CC BY 4.0, Fonts: SIL OFL 1.1, Code: MIT License) Copyright 2025 Fonticons, Inc.--><path d="M173.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6m-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3m44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9M252.8 8C114.1 8 8 113.3 8 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C436.2 457.8 504 362.9 504 252 504 113.3 391.5 8 252.8 8M105.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1m-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7m32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1m-11.4-14.7c-1.6 1-1.6 3.6 0 5.9s4.3 3.3 5.6 2.3c1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2"/></svg>
|
|
</div>
|
|
<div class="md-source__repository">
|
|
chopratejas/headroom
|
|
</div>
|
|
</a>
|
|
</div>
|
|
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href=".." class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Home
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--nested">
|
|
|
|
|
|
|
|
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_2" >
|
|
|
|
|
|
<label class="md-nav__link" for="__nav_2" id="__nav_2_label" tabindex="0">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Getting Started
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_2_label" aria-expanded="false">
|
|
<label class="md-nav__title" for="__nav_2">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
|
|
|
|
Getting Started
|
|
|
|
|
|
</label>
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../quickstart/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Quickstart
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../getting-started/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Installation
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--nested">
|
|
|
|
|
|
|
|
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_3" >
|
|
|
|
|
|
<label class="md-nav__link" for="__nav_3" id="__nav_3_label" tabindex="0">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
User Guide
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_3_label" aria-expanded="false">
|
|
<label class="md-nav__title" for="__nav_3">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
|
|
|
|
User Guide
|
|
|
|
|
|
</label>
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../proxy/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Proxy Server
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../compression/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Compression
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../image-compression/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Image Compression
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../ccr/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
CCR (Compress-Cache-Retrieve)
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../configuration/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Configuration
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--nested">
|
|
|
|
|
|
|
|
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_4" >
|
|
|
|
|
|
<label class="md-nav__link" for="__nav_4" id="__nav_4_label" tabindex="0">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Integrations
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_4_label" aria-expanded="false">
|
|
<label class="md-nav__title" for="__nav_4">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
|
|
|
|
Integrations
|
|
|
|
|
|
</label>
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../langchain/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
LangChain
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../agno/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Agno
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--active md-nav__item--section md-nav__item--nested">
|
|
|
|
|
|
|
|
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_5" checked>
|
|
|
|
|
|
<label class="md-nav__link" for="__nav_5" id="__nav_5_label" tabindex="">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Advanced
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_5_label" aria-expanded="true">
|
|
<label class="md-nav__title" for="__nav_5">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
|
|
|
|
Advanced
|
|
|
|
|
|
</label>
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../llmlingua/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
LLMLingua
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../text-compression/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Text Compression
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--active">
|
|
|
|
<input class="md-nav__toggle md-toggle" type="checkbox" id="__toc">
|
|
|
|
|
|
|
|
|
|
|
|
<label class="md-nav__link md-nav__link--active" for="__toc">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Transforms
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<a href="./" class="md-nav__link md-nav__link--active">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Transforms
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
|
|
|
|
|
|
<nav class="md-nav md-nav--secondary" aria-label="Table of contents">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<label class="md-nav__title" for="__toc">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
Table of contents
|
|
</label>
|
|
<ul class="md-nav__list" data-md-component="toc" data-md-scrollfix>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#smartcrusher" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
SmartCrusher
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="SmartCrusher">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#how-it-works" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
How It Works
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#what-gets-preserved" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
What Gets Preserved
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#cachealigner" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
CacheAligner
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="CacheAligner">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-problem" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Problem
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-solution" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Solution
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#cache-hit-improvement" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Cache Hit Improvement
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#rollingwindow" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
RollingWindow
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="RollingWindow">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-problem_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Problem
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-solution_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Solution
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#drop-priority" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Drop Priority
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#intelligentcontextmanager" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
IntelligentContextManager
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="IntelligentContextManager">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-problem_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Problem
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-solution_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Solution
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#how-scoring-works" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
How Scoring Works
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_3" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#strategy-selection" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Strategy Selection
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#toin-ccr-integration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
TOIN + CCR Integration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example-before-vs-after" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example: Before vs After
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#backwards-compatibility" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Backwards Compatibility
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#llmlinguacompressor-optional" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
LLMLinguaCompressor (Optional)
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="LLMLinguaCompressor (Optional)">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#when-to-use" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
When to Use
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#installation" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Installation
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_4" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#content-aware-rates" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Content-Aware Rates
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#memory-management" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Memory Management
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#proxy-integration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Proxy Integration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#codeawarecompressor-optional" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
CodeAwareCompressor (Optional)
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="CodeAwareCompressor (Optional)">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#when-to-use_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
When to Use
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#key-benefits" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Key Benefits
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#installation_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Installation
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_5" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#supported-languages" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Supported Languages
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#memory-management_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Memory Management
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#contentrouter" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
ContentRouter
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="ContentRouter">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#how-it-works_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
How It Works
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_6" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#compression-strategies" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Compression Strategies
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#content-detection" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Content Detection
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#toin-integration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
TOIN Integration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#transformpipeline" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
TransformPipeline
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="TransformPipeline">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#with-llmlingua-optional" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
With LLMLingua (Optional)
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#recommended-order" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Recommended Order
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#safety-guarantees" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Safety Guarantees
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../ARCHITECTURE/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Architecture
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--nested">
|
|
|
|
|
|
|
|
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_6" >
|
|
|
|
|
|
<label class="md-nav__link" for="__nav_6" id="__nav_6_label" tabindex="0">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Reference
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_6_label" aria-expanded="false">
|
|
<label class="md-nav__title" for="__nav_6">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
|
|
|
|
Reference
|
|
|
|
|
|
</label>
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../api/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
API
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../sdk/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
SDK
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../metrics/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Metrics
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../errors/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Errors
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../troubleshooting/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Troubleshooting
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--nested">
|
|
|
|
|
|
|
|
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_7" >
|
|
|
|
|
|
<label class="md-nav__link" for="__nav_7" id="__nav_7_label" tabindex="0">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Deployment
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_7_label" aria-expanded="false">
|
|
<label class="md-nav__title" for="__nav_7">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
|
|
|
|
Deployment
|
|
|
|
|
|
</label>
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../macos-deployment/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
macOS
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item md-nav__item--nested">
|
|
|
|
|
|
|
|
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_8" >
|
|
|
|
|
|
<label class="md-nav__link" for="__nav_8" id="__nav_8_label" tabindex="0">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Memory
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
<span class="md-nav__icon md-icon"></span>
|
|
</label>
|
|
|
|
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_8_label" aria-expanded="false">
|
|
<label class="md-nav__title" for="__nav_8">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
|
|
|
|
Memory
|
|
|
|
|
|
</label>
|
|
<ul class="md-nav__list" data-md-scrollfix>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<li class="md-nav__item">
|
|
<a href="../memory/" class="md-nav__link">
|
|
|
|
|
|
|
|
<span class="md-ellipsis">
|
|
|
|
|
|
Overview
|
|
|
|
|
|
|
|
</span>
|
|
|
|
|
|
|
|
</a>
|
|
</li>
|
|
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
|
|
|
|
</ul>
|
|
</nav>
|
|
</div>
|
|
</div>
|
|
</div>
|
|
|
|
|
|
|
|
<div class="md-sidebar md-sidebar--secondary" data-md-component="sidebar" data-md-type="toc" >
|
|
<div class="md-sidebar__scrollwrap">
|
|
<div class="md-sidebar__inner">
|
|
|
|
|
|
<nav class="md-nav md-nav--secondary" aria-label="Table of contents">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<label class="md-nav__title" for="__toc">
|
|
<span class="md-nav__icon md-icon"></span>
|
|
Table of contents
|
|
</label>
|
|
<ul class="md-nav__list" data-md-component="toc" data-md-scrollfix>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#smartcrusher" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
SmartCrusher
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="SmartCrusher">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#how-it-works" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
How It Works
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#what-gets-preserved" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
What Gets Preserved
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#cachealigner" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
CacheAligner
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="CacheAligner">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-problem" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Problem
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-solution" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Solution
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#cache-hit-improvement" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Cache Hit Improvement
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#rollingwindow" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
RollingWindow
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="RollingWindow">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-problem_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Problem
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-solution_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Solution
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#drop-priority" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Drop Priority
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#intelligentcontextmanager" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
IntelligentContextManager
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="IntelligentContextManager">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-problem_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Problem
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#the-solution_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
The Solution
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#how-scoring-works" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
How Scoring Works
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_3" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#strategy-selection" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Strategy Selection
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#toin-ccr-integration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
TOIN + CCR Integration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example-before-vs-after" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example: Before vs After
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#backwards-compatibility" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Backwards Compatibility
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#llmlinguacompressor-optional" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
LLMLinguaCompressor (Optional)
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="LLMLinguaCompressor (Optional)">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#when-to-use" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
When to Use
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#installation" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Installation
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_4" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#content-aware-rates" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Content-Aware Rates
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#memory-management" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Memory Management
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#proxy-integration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Proxy Integration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#codeawarecompressor-optional" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
CodeAwareCompressor (Optional)
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="CodeAwareCompressor (Optional)">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#when-to-use_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
When to Use
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#key-benefits" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Key Benefits
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#installation_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Installation
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_5" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#supported-languages" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Supported Languages
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#memory-management_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Memory Management
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#contentrouter" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
ContentRouter
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="ContentRouter">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#how-it-works_1" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
How It Works
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#configuration_6" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Configuration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#example_2" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Example
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#compression-strategies" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Compression Strategies
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#content-detection" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Content Detection
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#toin-integration" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
TOIN Integration
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#transformpipeline" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
TransformPipeline
|
|
|
|
</span>
|
|
</a>
|
|
|
|
<nav class="md-nav" aria-label="TransformPipeline">
|
|
<ul class="md-nav__list">
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#with-llmlingua-optional" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
With LLMLingua (Optional)
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#recommended-order" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Recommended Order
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
</nav>
|
|
|
|
</li>
|
|
|
|
<li class="md-nav__item">
|
|
<a href="#safety-guarantees" class="md-nav__link">
|
|
<span class="md-ellipsis">
|
|
|
|
Safety Guarantees
|
|
|
|
</span>
|
|
</a>
|
|
|
|
</li>
|
|
|
|
</ul>
|
|
|
|
</nav>
|
|
</div>
|
|
</div>
|
|
</div>
|
|
|
|
|
|
|
|
<div class="md-content" data-md-component="content">
|
|
|
|
<article class="md-content__inner md-typeset">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<h1 id="transform-reference">Transform Reference<a class="headerlink" href="#transform-reference" title="Permanent link">¶</a></h1>
|
|
<p>Headroom provides several transforms that work together to optimize LLM context.</p>
|
|
<h2 id="smartcrusher">SmartCrusher<a class="headerlink" href="#smartcrusher" title="Permanent link">¶</a></h2>
|
|
<p>Statistical compression for JSON tool outputs.</p>
|
|
<h3 id="how-it-works">How It Works<a class="headerlink" href="#how-it-works" title="Permanent link">¶</a></h3>
|
|
<p>SmartCrusher analyzes JSON arrays and selectively keeps important items:</p>
|
|
<ol>
|
|
<li><strong>First/Last items</strong> - Context for pagination and recency</li>
|
|
<li><strong>Error items</strong> - 100% preservation of error states</li>
|
|
<li><strong>Anomalies</strong> - Statistical outliers (> 2 std dev from mean)</li>
|
|
<li><strong>Relevant items</strong> - Matches to user's query via BM25/embeddings</li>
|
|
<li><strong>Change points</strong> - Significant transitions in data</li>
|
|
</ol>
|
|
<h3 id="configuration">Configuration<a class="headerlink" href="#configuration" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-0-1"><a id="__codelineno-0-1" name="__codelineno-0-1" href="#__codelineno-0-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom</span><span class="w"> </span><span class="kn">import</span> <span class="n">SmartCrusherConfig</span>
|
|
</span><span id="__span-0-2"><a id="__codelineno-0-2" name="__codelineno-0-2" href="#__codelineno-0-2"></a>
|
|
</span><span id="__span-0-3"><a id="__codelineno-0-3" name="__codelineno-0-3" href="#__codelineno-0-3"></a><span class="n">config</span> <span class="o">=</span> <span class="n">SmartCrusherConfig</span><span class="p">(</span>
|
|
</span><span id="__span-0-4"><a id="__codelineno-0-4" name="__codelineno-0-4" href="#__codelineno-0-4"></a> <span class="n">min_tokens_to_crush</span><span class="o">=</span><span class="mi">200</span><span class="p">,</span> <span class="c1"># Only compress if > 200 tokens</span>
|
|
</span><span id="__span-0-5"><a id="__codelineno-0-5" name="__codelineno-0-5" href="#__codelineno-0-5"></a> <span class="n">max_items_after_crush</span><span class="o">=</span><span class="mi">50</span><span class="p">,</span> <span class="c1"># Keep at most 50 items</span>
|
|
</span><span id="__span-0-6"><a id="__codelineno-0-6" name="__codelineno-0-6" href="#__codelineno-0-6"></a> <span class="n">keep_first</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span> <span class="c1"># Always keep first 3 items</span>
|
|
</span><span id="__span-0-7"><a id="__codelineno-0-7" name="__codelineno-0-7" href="#__codelineno-0-7"></a> <span class="n">keep_last</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="c1"># Always keep last 2 items</span>
|
|
</span><span id="__span-0-8"><a id="__codelineno-0-8" name="__codelineno-0-8" href="#__codelineno-0-8"></a> <span class="n">relevance_threshold</span><span class="o">=</span><span class="mf">0.3</span><span class="p">,</span> <span class="c1"># Keep items with relevance > 0.3</span>
|
|
</span><span id="__span-0-9"><a id="__codelineno-0-9" name="__codelineno-0-9" href="#__codelineno-0-9"></a> <span class="n">anomaly_std_threshold</span><span class="o">=</span><span class="mf">2.0</span><span class="p">,</span> <span class="c1"># Keep items > 2 std dev from mean</span>
|
|
</span><span id="__span-0-10"><a id="__codelineno-0-10" name="__codelineno-0-10" href="#__codelineno-0-10"></a> <span class="n">preserve_errors</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Always keep error items</span>
|
|
</span><span id="__span-0-11"><a id="__codelineno-0-11" name="__codelineno-0-11" href="#__codelineno-0-11"></a><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="example">Example<a class="headerlink" href="#example" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-1-1"><a id="__codelineno-1-1" name="__codelineno-1-1" href="#__codelineno-1-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom</span><span class="w"> </span><span class="kn">import</span> <span class="n">SmartCrusher</span>
|
|
</span><span id="__span-1-2"><a id="__codelineno-1-2" name="__codelineno-1-2" href="#__codelineno-1-2"></a>
|
|
</span><span id="__span-1-3"><a id="__codelineno-1-3" name="__codelineno-1-3" href="#__codelineno-1-3"></a><span class="n">crusher</span> <span class="o">=</span> <span class="n">SmartCrusher</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
|
</span><span id="__span-1-4"><a id="__codelineno-1-4" name="__codelineno-1-4" href="#__codelineno-1-4"></a>
|
|
</span><span id="__span-1-5"><a id="__codelineno-1-5" name="__codelineno-1-5" href="#__codelineno-1-5"></a><span class="c1"># Before: 1000 search results (45,000 tokens)</span>
|
|
</span><span id="__span-1-6"><a id="__codelineno-1-6" name="__codelineno-1-6" href="#__codelineno-1-6"></a><span class="n">tool_output</span> <span class="o">=</span> <span class="p">{</span><span class="s2">"results"</span><span class="p">:</span> <span class="p">[</span><span class="o">..</span><span class="mf">.1000</span> <span class="n">items</span><span class="o">...</span><span class="p">]}</span>
|
|
</span><span id="__span-1-7"><a id="__codelineno-1-7" name="__codelineno-1-7" href="#__codelineno-1-7"></a>
|
|
</span><span id="__span-1-8"><a id="__codelineno-1-8" name="__codelineno-1-8" href="#__codelineno-1-8"></a><span class="c1"># After: ~50 important items (4,500 tokens) - 90% reduction</span>
|
|
</span><span id="__span-1-9"><a id="__codelineno-1-9" name="__codelineno-1-9" href="#__codelineno-1-9"></a><span class="n">compressed</span> <span class="o">=</span> <span class="n">crusher</span><span class="o">.</span><span class="n">crush</span><span class="p">(</span><span class="n">tool_output</span><span class="p">,</span> <span class="n">query</span><span class="o">=</span><span class="s2">"user's question"</span><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="what-gets-preserved">What Gets Preserved<a class="headerlink" href="#what-gets-preserved" title="Permanent link">¶</a></h3>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Category</th>
|
|
<th>Preserved</th>
|
|
<th>Why</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>Errors</td>
|
|
<td>100%</td>
|
|
<td>Critical for debugging</td>
|
|
</tr>
|
|
<tr>
|
|
<td>First N</td>
|
|
<td>100%</td>
|
|
<td>Context/pagination</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Last N</td>
|
|
<td>100%</td>
|
|
<td>Recency</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Anomalies</td>
|
|
<td>All</td>
|
|
<td>Unusual values matter</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Relevant</td>
|
|
<td>Top K</td>
|
|
<td>Match user's query</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Others</td>
|
|
<td>Sampled</td>
|
|
<td>Statistical representation</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<hr />
|
|
<h2 id="cachealigner">CacheAligner<a class="headerlink" href="#cachealigner" title="Permanent link">¶</a></h2>
|
|
<p>Prefix stabilization for improved cache hit rates.</p>
|
|
<h3 id="the-problem">The Problem<a class="headerlink" href="#the-problem" title="Permanent link">¶</a></h3>
|
|
<p>LLM providers cache request prefixes. But dynamic content breaks caching:</p>
|
|
<div class="language-text highlight"><pre><span></span><code><span id="__span-2-1"><a id="__codelineno-2-1" name="__codelineno-2-1" href="#__codelineno-2-1"></a>"You are helpful. Today is January 7, 2025." # Changes daily = no cache
|
|
</span></code></pre></div>
|
|
<h3 id="the-solution">The Solution<a class="headerlink" href="#the-solution" title="Permanent link">¶</a></h3>
|
|
<p>CacheAligner extracts dynamic content to stabilize the prefix:</p>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-3-1"><a id="__codelineno-3-1" name="__codelineno-3-1" href="#__codelineno-3-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom</span><span class="w"> </span><span class="kn">import</span> <span class="n">CacheAligner</span>
|
|
</span><span id="__span-3-2"><a id="__codelineno-3-2" name="__codelineno-3-2" href="#__codelineno-3-2"></a>
|
|
</span><span id="__span-3-3"><a id="__codelineno-3-3" name="__codelineno-3-3" href="#__codelineno-3-3"></a><span class="n">aligner</span> <span class="o">=</span> <span class="n">CacheAligner</span><span class="p">()</span>
|
|
</span><span id="__span-3-4"><a id="__codelineno-3-4" name="__codelineno-3-4" href="#__codelineno-3-4"></a><span class="n">result</span> <span class="o">=</span> <span class="n">aligner</span><span class="o">.</span><span class="n">align</span><span class="p">(</span><span class="n">messages</span><span class="p">)</span>
|
|
</span><span id="__span-3-5"><a id="__codelineno-3-5" name="__codelineno-3-5" href="#__codelineno-3-5"></a>
|
|
</span><span id="__span-3-6"><a id="__codelineno-3-6" name="__codelineno-3-6" href="#__codelineno-3-6"></a><span class="c1"># Static prefix (cacheable):</span>
|
|
</span><span id="__span-3-7"><a id="__codelineno-3-7" name="__codelineno-3-7" href="#__codelineno-3-7"></a><span class="c1"># "You are helpful."</span>
|
|
</span><span id="__span-3-8"><a id="__codelineno-3-8" name="__codelineno-3-8" href="#__codelineno-3-8"></a>
|
|
</span><span id="__span-3-9"><a id="__codelineno-3-9" name="__codelineno-3-9" href="#__codelineno-3-9"></a><span class="c1"># Dynamic content moved to end:</span>
|
|
</span><span id="__span-3-10"><a id="__codelineno-3-10" name="__codelineno-3-10" href="#__codelineno-3-10"></a><span class="c1"># [Current date context]</span>
|
|
</span></code></pre></div>
|
|
<h3 id="configuration_1">Configuration<a class="headerlink" href="#configuration_1" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-4-1"><a id="__codelineno-4-1" name="__codelineno-4-1" href="#__codelineno-4-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom</span><span class="w"> </span><span class="kn">import</span> <span class="n">CacheAlignerConfig</span>
|
|
</span><span id="__span-4-2"><a id="__codelineno-4-2" name="__codelineno-4-2" href="#__codelineno-4-2"></a>
|
|
</span><span id="__span-4-3"><a id="__codelineno-4-3" name="__codelineno-4-3" href="#__codelineno-4-3"></a><span class="n">config</span> <span class="o">=</span> <span class="n">CacheAlignerConfig</span><span class="p">(</span>
|
|
</span><span id="__span-4-4"><a id="__codelineno-4-4" name="__codelineno-4-4" href="#__codelineno-4-4"></a> <span class="n">extract_dates</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Move dates to dynamic section</span>
|
|
</span><span id="__span-4-5"><a id="__codelineno-4-5" name="__codelineno-4-5" href="#__codelineno-4-5"></a> <span class="n">normalize_whitespace</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Consistent spacing</span>
|
|
</span><span id="__span-4-6"><a id="__codelineno-4-6" name="__codelineno-4-6" href="#__codelineno-4-6"></a> <span class="n">stable_prefix_min_tokens</span><span class="o">=</span><span class="mi">100</span><span class="p">,</span> <span class="c1"># Min prefix size for alignment</span>
|
|
</span><span id="__span-4-7"><a id="__codelineno-4-7" name="__codelineno-4-7" href="#__codelineno-4-7"></a><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="cache-hit-improvement">Cache Hit Improvement<a class="headerlink" href="#cache-hit-improvement" title="Permanent link">¶</a></h3>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Scenario</th>
|
|
<th>Before</th>
|
|
<th>After</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>Daily date in prompt</td>
|
|
<td>0% hits</td>
|
|
<td>~95% hits</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Dynamic user context</td>
|
|
<td>~10% hits</td>
|
|
<td>~80% hits</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Consistent prompts</td>
|
|
<td>~90% hits</td>
|
|
<td>~95% hits</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<hr />
|
|
<h2 id="rollingwindow">RollingWindow<a class="headerlink" href="#rollingwindow" title="Permanent link">¶</a></h2>
|
|
<p>Context management within token limits.</p>
|
|
<h3 id="the-problem_1">The Problem<a class="headerlink" href="#the-problem_1" title="Permanent link">¶</a></h3>
|
|
<p>Long conversations exceed context limits. Naive truncation breaks tool calls:</p>
|
|
<div class="language-text highlight"><pre><span></span><code><span id="__span-5-1"><a id="__codelineno-5-1" name="__codelineno-5-1" href="#__codelineno-5-1"></a>[tool_call: search] # Kept
|
|
</span><span id="__span-5-2"><a id="__codelineno-5-2" name="__codelineno-5-2" href="#__codelineno-5-2"></a>[tool_result: ...] # Dropped = orphaned call!
|
|
</span></code></pre></div>
|
|
<h3 id="the-solution_1">The Solution<a class="headerlink" href="#the-solution_1" title="Permanent link">¶</a></h3>
|
|
<p>RollingWindow drops complete tool units, preserving pairs:</p>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-6-1"><a id="__codelineno-6-1" name="__codelineno-6-1" href="#__codelineno-6-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom</span><span class="w"> </span><span class="kn">import</span> <span class="n">RollingWindow</span>
|
|
</span><span id="__span-6-2"><a id="__codelineno-6-2" name="__codelineno-6-2" href="#__codelineno-6-2"></a>
|
|
</span><span id="__span-6-3"><a id="__codelineno-6-3" name="__codelineno-6-3" href="#__codelineno-6-3"></a><span class="n">window</span> <span class="o">=</span> <span class="n">RollingWindow</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
|
</span><span id="__span-6-4"><a id="__codelineno-6-4" name="__codelineno-6-4" href="#__codelineno-6-4"></a><span class="n">result</span> <span class="o">=</span> <span class="n">window</span><span class="o">.</span><span class="n">apply</span><span class="p">(</span><span class="n">messages</span><span class="p">,</span> <span class="n">max_tokens</span><span class="o">=</span><span class="mi">100000</span><span class="p">)</span>
|
|
</span><span id="__span-6-5"><a id="__codelineno-6-5" name="__codelineno-6-5" href="#__codelineno-6-5"></a>
|
|
</span><span id="__span-6-6"><a id="__codelineno-6-6" name="__codelineno-6-6" href="#__codelineno-6-6"></a><span class="c1"># Guarantees:</span>
|
|
</span><span id="__span-6-7"><a id="__codelineno-6-7" name="__codelineno-6-7" href="#__codelineno-6-7"></a><span class="c1"># 1. Tool calls paired with results</span>
|
|
</span><span id="__span-6-8"><a id="__codelineno-6-8" name="__codelineno-6-8" href="#__codelineno-6-8"></a><span class="c1"># 2. System prompt preserved</span>
|
|
</span><span id="__span-6-9"><a id="__codelineno-6-9" name="__codelineno-6-9" href="#__codelineno-6-9"></a><span class="c1"># 3. Recent turns kept</span>
|
|
</span><span id="__span-6-10"><a id="__codelineno-6-10" name="__codelineno-6-10" href="#__codelineno-6-10"></a><span class="c1"># 4. Oldest tool outputs dropped first</span>
|
|
</span></code></pre></div>
|
|
<h3 id="configuration_2">Configuration<a class="headerlink" href="#configuration_2" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-7-1"><a id="__codelineno-7-1" name="__codelineno-7-1" href="#__codelineno-7-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom</span><span class="w"> </span><span class="kn">import</span> <span class="n">RollingWindowConfig</span>
|
|
</span><span id="__span-7-2"><a id="__codelineno-7-2" name="__codelineno-7-2" href="#__codelineno-7-2"></a>
|
|
</span><span id="__span-7-3"><a id="__codelineno-7-3" name="__codelineno-7-3" href="#__codelineno-7-3"></a><span class="n">config</span> <span class="o">=</span> <span class="n">RollingWindowConfig</span><span class="p">(</span>
|
|
</span><span id="__span-7-4"><a id="__codelineno-7-4" name="__codelineno-7-4" href="#__codelineno-7-4"></a> <span class="n">max_tokens</span><span class="o">=</span><span class="mi">100000</span><span class="p">,</span> <span class="c1"># Target token limit</span>
|
|
</span><span id="__span-7-5"><a id="__codelineno-7-5" name="__codelineno-7-5" href="#__codelineno-7-5"></a> <span class="n">preserve_system</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Always keep system prompt</span>
|
|
</span><span id="__span-7-6"><a id="__codelineno-7-6" name="__codelineno-7-6" href="#__codelineno-7-6"></a> <span class="n">preserve_recent_turns</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="c1"># Keep last 5 user/assistant turns</span>
|
|
</span><span id="__span-7-7"><a id="__codelineno-7-7" name="__codelineno-7-7" href="#__codelineno-7-7"></a> <span class="n">drop_oldest_first</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Remove oldest tool outputs</span>
|
|
</span><span id="__span-7-8"><a id="__codelineno-7-8" name="__codelineno-7-8" href="#__codelineno-7-8"></a><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="drop-priority">Drop Priority<a class="headerlink" href="#drop-priority" title="Permanent link">¶</a></h3>
|
|
<ol>
|
|
<li><strong>Oldest tool outputs</strong> - First to go</li>
|
|
<li><strong>Old assistant messages</strong> - Summary preserved</li>
|
|
<li><strong>Old user messages</strong> - Only if necessary</li>
|
|
<li><strong>Never dropped</strong>: System prompt, recent turns, active tool pairs</li>
|
|
</ol>
|
|
<blockquote>
|
|
<p><strong>Note:</strong> For more intelligent context management based on semantic importance rather than just position, see <a href="#intelligentcontextmanager">IntelligentContextManager</a> below.</p>
|
|
</blockquote>
|
|
<hr />
|
|
<h2 id="intelligentcontextmanager">IntelligentContextManager<a class="headerlink" href="#intelligentcontextmanager" title="Permanent link">¶</a></h2>
|
|
<p>Semantic-aware context management with TOIN-learned importance scoring.</p>
|
|
<h3 id="the-problem_2">The Problem<a class="headerlink" href="#the-problem_2" title="Permanent link">¶</a></h3>
|
|
<p>RollingWindow drops messages by position (oldest first), but position doesn't equal importance:</p>
|
|
<ul>
|
|
<li>An error message from turn 3 might be critical</li>
|
|
<li>A verbose success response from turn 10 might be expendable</li>
|
|
<li>Messages referenced by later turns should be preserved</li>
|
|
</ul>
|
|
<h3 id="the-solution_2">The Solution<a class="headerlink" href="#the-solution_2" title="Permanent link">¶</a></h3>
|
|
<p>IntelligentContextManager uses multi-factor importance scoring:</p>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-8-1"><a id="__codelineno-8-1" name="__codelineno-8-1" href="#__codelineno-8-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">IntelligentContextManager</span><span class="p">,</span> <span class="n">IntelligentContextConfig</span>
|
|
</span><span id="__span-8-2"><a id="__codelineno-8-2" name="__codelineno-8-2" href="#__codelineno-8-2"></a>
|
|
</span><span id="__span-8-3"><a id="__codelineno-8-3" name="__codelineno-8-3" href="#__codelineno-8-3"></a><span class="n">manager</span> <span class="o">=</span> <span class="n">IntelligentContextManager</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
|
</span><span id="__span-8-4"><a id="__codelineno-8-4" name="__codelineno-8-4" href="#__codelineno-8-4"></a><span class="n">result</span> <span class="o">=</span> <span class="n">manager</span><span class="o">.</span><span class="n">apply</span><span class="p">(</span><span class="n">messages</span><span class="p">,</span> <span class="n">tokenizer</span><span class="p">,</span> <span class="n">model_limit</span><span class="o">=</span><span class="mi">128000</span><span class="p">)</span>
|
|
</span><span id="__span-8-5"><a id="__codelineno-8-5" name="__codelineno-8-5" href="#__codelineno-8-5"></a>
|
|
</span><span id="__span-8-6"><a id="__codelineno-8-6" name="__codelineno-8-6" href="#__codelineno-8-6"></a><span class="c1"># Guarantees:</span>
|
|
</span><span id="__span-8-7"><a id="__codelineno-8-7" name="__codelineno-8-7" href="#__codelineno-8-7"></a><span class="c1"># 1. System messages never dropped (configurable)</span>
|
|
</span><span id="__span-8-8"><a id="__codelineno-8-8" name="__codelineno-8-8" href="#__codelineno-8-8"></a><span class="c1"># 2. Last N turns always protected</span>
|
|
</span><span id="__span-8-9"><a id="__codelineno-8-9" name="__codelineno-8-9" href="#__codelineno-8-9"></a><span class="c1"># 3. Tool calls/responses dropped atomically</span>
|
|
</span><span id="__span-8-10"><a id="__codelineno-8-10" name="__codelineno-8-10" href="#__codelineno-8-10"></a><span class="c1"># 4. Drops by importance score, not just position</span>
|
|
</span></code></pre></div>
|
|
<h3 id="how-scoring-works">How Scoring Works<a class="headerlink" href="#how-scoring-works" title="Permanent link">¶</a></h3>
|
|
<p>Messages are scored on multiple factors (all learned, no hardcodes):</p>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Factor</th>
|
|
<th>Weight</th>
|
|
<th>Description</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>Recency</td>
|
|
<td>20%</td>
|
|
<td>Exponential decay from conversation end</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Semantic Similarity</td>
|
|
<td>20%</td>
|
|
<td>Embedding similarity to recent context</td>
|
|
</tr>
|
|
<tr>
|
|
<td>TOIN Importance</td>
|
|
<td>25%</td>
|
|
<td>Learned from retrieval patterns</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Error Indicators</td>
|
|
<td>15%</td>
|
|
<td>TOIN-learned error field detection</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Forward References</td>
|
|
<td>15%</td>
|
|
<td>Messages referenced by later messages</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Token Density</td>
|
|
<td>5%</td>
|
|
<td>Information density (unique/total tokens)</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<p><strong>Key principle:</strong> No hardcoded patterns. Error detection uses TOIN's <code>field_semantics.inferred_type == "error_indicator"</code>, not keyword matching.</p>
|
|
<h3 id="configuration_3">Configuration<a class="headerlink" href="#configuration_3" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-9-1"><a id="__codelineno-9-1" name="__codelineno-9-1" href="#__codelineno-9-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">IntelligentContextManager</span>
|
|
</span><span id="__span-9-2"><a id="__codelineno-9-2" name="__codelineno-9-2" href="#__codelineno-9-2"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.config</span><span class="w"> </span><span class="kn">import</span> <span class="n">IntelligentContextConfig</span><span class="p">,</span> <span class="n">ScoringWeights</span>
|
|
</span><span id="__span-9-3"><a id="__codelineno-9-3" name="__codelineno-9-3" href="#__codelineno-9-3"></a>
|
|
</span><span id="__span-9-4"><a id="__codelineno-9-4" name="__codelineno-9-4" href="#__codelineno-9-4"></a><span class="c1"># Custom scoring weights</span>
|
|
</span><span id="__span-9-5"><a id="__codelineno-9-5" name="__codelineno-9-5" href="#__codelineno-9-5"></a><span class="n">weights</span> <span class="o">=</span> <span class="n">ScoringWeights</span><span class="p">(</span>
|
|
</span><span id="__span-9-6"><a id="__codelineno-9-6" name="__codelineno-9-6" href="#__codelineno-9-6"></a> <span class="n">recency</span><span class="o">=</span><span class="mf">0.20</span><span class="p">,</span>
|
|
</span><span id="__span-9-7"><a id="__codelineno-9-7" name="__codelineno-9-7" href="#__codelineno-9-7"></a> <span class="n">semantic_similarity</span><span class="o">=</span><span class="mf">0.20</span><span class="p">,</span>
|
|
</span><span id="__span-9-8"><a id="__codelineno-9-8" name="__codelineno-9-8" href="#__codelineno-9-8"></a> <span class="n">toin_importance</span><span class="o">=</span><span class="mf">0.25</span><span class="p">,</span>
|
|
</span><span id="__span-9-9"><a id="__codelineno-9-9" name="__codelineno-9-9" href="#__codelineno-9-9"></a> <span class="n">error_indicator</span><span class="o">=</span><span class="mf">0.15</span><span class="p">,</span>
|
|
</span><span id="__span-9-10"><a id="__codelineno-9-10" name="__codelineno-9-10" href="#__codelineno-9-10"></a> <span class="n">forward_reference</span><span class="o">=</span><span class="mf">0.15</span><span class="p">,</span>
|
|
</span><span id="__span-9-11"><a id="__codelineno-9-11" name="__codelineno-9-11" href="#__codelineno-9-11"></a> <span class="n">token_density</span><span class="o">=</span><span class="mf">0.05</span><span class="p">,</span>
|
|
</span><span id="__span-9-12"><a id="__codelineno-9-12" name="__codelineno-9-12" href="#__codelineno-9-12"></a><span class="p">)</span>
|
|
</span><span id="__span-9-13"><a id="__codelineno-9-13" name="__codelineno-9-13" href="#__codelineno-9-13"></a>
|
|
</span><span id="__span-9-14"><a id="__codelineno-9-14" name="__codelineno-9-14" href="#__codelineno-9-14"></a><span class="n">config</span> <span class="o">=</span> <span class="n">IntelligentContextConfig</span><span class="p">(</span>
|
|
</span><span id="__span-9-15"><a id="__codelineno-9-15" name="__codelineno-9-15" href="#__codelineno-9-15"></a> <span class="n">enabled</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
</span><span id="__span-9-16"><a id="__codelineno-9-16" name="__codelineno-9-16" href="#__codelineno-9-16"></a> <span class="n">keep_system</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Never drop system messages</span>
|
|
</span><span id="__span-9-17"><a id="__codelineno-9-17" name="__codelineno-9-17" href="#__codelineno-9-17"></a> <span class="n">keep_last_turns</span><span class="o">=</span><span class="mi">2</span><span class="p">,</span> <span class="c1"># Protect last N user turns</span>
|
|
</span><span id="__span-9-18"><a id="__codelineno-9-18" name="__codelineno-9-18" href="#__codelineno-9-18"></a> <span class="n">output_buffer_tokens</span><span class="o">=</span><span class="mi">4000</span><span class="p">,</span> <span class="c1"># Reserve for model output</span>
|
|
</span><span id="__span-9-19"><a id="__codelineno-9-19" name="__codelineno-9-19" href="#__codelineno-9-19"></a> <span class="n">use_importance_scoring</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Enable semantic scoring</span>
|
|
</span><span id="__span-9-20"><a id="__codelineno-9-20" name="__codelineno-9-20" href="#__codelineno-9-20"></a> <span class="n">scoring_weights</span><span class="o">=</span><span class="n">weights</span><span class="p">,</span> <span class="c1"># Custom weights</span>
|
|
</span><span id="__span-9-21"><a id="__codelineno-9-21" name="__codelineno-9-21" href="#__codelineno-9-21"></a> <span class="n">toin_integration</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Use TOIN patterns</span>
|
|
</span><span id="__span-9-22"><a id="__codelineno-9-22" name="__codelineno-9-22" href="#__codelineno-9-22"></a> <span class="n">recency_decay_rate</span><span class="o">=</span><span class="mf">0.1</span><span class="p">,</span> <span class="c1"># Exponential decay lambda</span>
|
|
</span><span id="__span-9-23"><a id="__codelineno-9-23" name="__codelineno-9-23" href="#__codelineno-9-23"></a> <span class="n">compress_threshold</span><span class="o">=</span><span class="mf">0.1</span><span class="p">,</span> <span class="c1"># Try compression first if <10% over</span>
|
|
</span><span id="__span-9-24"><a id="__codelineno-9-24" name="__codelineno-9-24" href="#__codelineno-9-24"></a><span class="p">)</span>
|
|
</span><span id="__span-9-25"><a id="__codelineno-9-25" name="__codelineno-9-25" href="#__codelineno-9-25"></a>
|
|
</span><span id="__span-9-26"><a id="__codelineno-9-26" name="__codelineno-9-26" href="#__codelineno-9-26"></a><span class="n">manager</span> <span class="o">=</span> <span class="n">IntelligentContextManager</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="strategy-selection">Strategy Selection<a class="headerlink" href="#strategy-selection" title="Permanent link">¶</a></h3>
|
|
<p>Based on how much over budget you are:</p>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Overage</th>
|
|
<th>Strategy</th>
|
|
<th>Action</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>Under budget</td>
|
|
<td>NONE</td>
|
|
<td>No action needed</td>
|
|
</tr>
|
|
<tr>
|
|
<td>< 10% over</td>
|
|
<td>COMPRESS_FIRST</td>
|
|
<td>Try deeper compression</td>
|
|
</tr>
|
|
<tr>
|
|
<td>>= 10% over</td>
|
|
<td>DROP_BY_SCORE</td>
|
|
<td>Drop lowest-scored messages</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<h3 id="toin-ccr-integration">TOIN + CCR Integration<a class="headerlink" href="#toin-ccr-integration" title="Permanent link">¶</a></h3>
|
|
<p>IntelligentContextManager is a <strong>message-level compressor</strong>. Just like SmartCrusher compresses items in a JSON array, IntelligentContext "compresses" messages in a conversation by dropping low-value ones.</p>
|
|
<p><strong>Bidirectional TOIN integration:</strong></p>
|
|
<ol>
|
|
<li><strong>Scoring uses TOIN patterns</strong>: Learned retrieval rates and field semantics inform importance scores</li>
|
|
<li><strong>Drops are recorded to TOIN</strong>: When messages are dropped, TOIN learns the pattern</li>
|
|
<li><strong>CCR stores originals</strong>: Dropped messages are stored in CCR for potential retrieval</li>
|
|
<li><strong>Retrievals feed back to TOIN</strong>: If users retrieve dropped messages, TOIN learns to score those patterns higher</li>
|
|
</ol>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-10-1"><a id="__codelineno-10-1" name="__codelineno-10-1" href="#__codelineno-10-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.telemetry</span><span class="w"> </span><span class="kn">import</span> <span class="n">get_toin</span>
|
|
</span><span id="__span-10-2"><a id="__codelineno-10-2" name="__codelineno-10-2" href="#__codelineno-10-2"></a>
|
|
</span><span id="__span-10-3"><a id="__codelineno-10-3" name="__codelineno-10-3" href="#__codelineno-10-3"></a><span class="n">toin</span> <span class="o">=</span> <span class="n">get_toin</span><span class="p">()</span>
|
|
</span><span id="__span-10-4"><a id="__codelineno-10-4" name="__codelineno-10-4" href="#__codelineno-10-4"></a><span class="n">manager</span> <span class="o">=</span> <span class="n">IntelligentContextManager</span><span class="p">(</span><span class="n">config</span><span class="p">,</span> <span class="n">toin</span><span class="o">=</span><span class="n">toin</span><span class="p">)</span>
|
|
</span><span id="__span-10-5"><a id="__codelineno-10-5" name="__codelineno-10-5" href="#__codelineno-10-5"></a>
|
|
</span><span id="__span-10-6"><a id="__codelineno-10-6" name="__codelineno-10-6" href="#__codelineno-10-6"></a><span class="c1"># TOIN provides (for scoring):</span>
|
|
</span><span id="__span-10-7"><a id="__codelineno-10-7" name="__codelineno-10-7" href="#__codelineno-10-7"></a><span class="c1"># - retrieval_rate: How often this message pattern is retrieved (high = important)</span>
|
|
</span><span id="__span-10-8"><a id="__codelineno-10-8" name="__codelineno-10-8" href="#__codelineno-10-8"></a><span class="c1"># - field_semantics: Learned field types (error_indicator, identifier, etc.)</span>
|
|
</span><span id="__span-10-9"><a id="__codelineno-10-9" name="__codelineno-10-9" href="#__codelineno-10-9"></a><span class="c1"># - commonly_retrieved_fields: Fields that users frequently need</span>
|
|
</span><span id="__span-10-10"><a id="__codelineno-10-10" name="__codelineno-10-10" href="#__codelineno-10-10"></a>
|
|
</span><span id="__span-10-11"><a id="__codelineno-10-11" name="__codelineno-10-11" href="#__codelineno-10-11"></a><span class="c1"># TOIN receives (from drops):</span>
|
|
</span><span id="__span-10-12"><a id="__codelineno-10-12" name="__codelineno-10-12" href="#__codelineno-10-12"></a><span class="c1"># - Message pattern signatures (role counts, has_tools, has_errors)</span>
|
|
</span><span id="__span-10-13"><a id="__codelineno-10-13" name="__codelineno-10-13" href="#__codelineno-10-13"></a><span class="c1"># - Token counts (original vs marker size)</span>
|
|
</span><span id="__span-10-14"><a id="__codelineno-10-14" name="__codelineno-10-14" href="#__codelineno-10-14"></a><span class="c1"># - Retrieval feedback when users access CCR</span>
|
|
</span></code></pre></div>
|
|
<p><strong>What this means:</strong>
|
|
- When you drop a message pattern and users frequently retrieve it, TOIN learns to score it higher next time
|
|
- When you drop a pattern and no one retrieves it, that confirms it was safe to drop
|
|
- The feedback loop improves drop decisions across all users, not just in one session</p>
|
|
<h3 id="example-before-vs-after">Example: Before vs After<a class="headerlink" href="#example-before-vs-after" title="Permanent link">¶</a></h3>
|
|
<p><strong>RollingWindow (position-based):</strong>
|
|
<div class="language-text highlight"><pre><span></span><code><span id="__span-11-1"><a id="__codelineno-11-1" name="__codelineno-11-1" href="#__codelineno-11-1"></a>Messages: [sys, user1, asst1, user2, asst2_error, user3, asst3, user4, asst4]
|
|
</span><span id="__span-11-2"><a id="__codelineno-11-2" name="__codelineno-11-2" href="#__codelineno-11-2"></a>Over budget by 3 messages.
|
|
</span><span id="__span-11-3"><a id="__codelineno-11-3" name="__codelineno-11-3" href="#__codelineno-11-3"></a>Drops: user1, asst1, user2 (oldest first)
|
|
</span><span id="__span-11-4"><a id="__codelineno-11-4" name="__codelineno-11-4" href="#__codelineno-11-4"></a>Result: Loses context, keeps verbose asst3
|
|
</span></code></pre></div></p>
|
|
<p><strong>IntelligentContextManager (score-based):</strong>
|
|
<div class="language-text highlight"><pre><span></span><code><span id="__span-12-1"><a id="__codelineno-12-1" name="__codelineno-12-1" href="#__codelineno-12-1"></a>Messages scored:
|
|
</span><span id="__span-12-2"><a id="__codelineno-12-2" name="__codelineno-12-2" href="#__codelineno-12-2"></a> - asst2_error: 0.85 (TOIN learned error indicator)
|
|
</span><span id="__span-12-3"><a id="__codelineno-12-3" name="__codelineno-12-3" href="#__codelineno-12-3"></a> - asst1: 0.45 (old, low density)
|
|
</span><span id="__span-12-4"><a id="__codelineno-12-4" name="__codelineno-12-4" href="#__codelineno-12-4"></a> - asst3: 0.40 (verbose, low unique tokens)
|
|
</span><span id="__span-12-5"><a id="__codelineno-12-5" name="__codelineno-12-5" href="#__codelineno-12-5"></a>
|
|
</span><span id="__span-12-6"><a id="__codelineno-12-6" name="__codelineno-12-6" href="#__codelineno-12-6"></a>Drops: asst1, asst3, user1 (lowest scores)
|
|
</span><span id="__span-12-7"><a id="__codelineno-12-7" name="__codelineno-12-7" href="#__codelineno-12-7"></a>Result: Preserves critical error message
|
|
</span></code></pre></div></p>
|
|
<h3 id="backwards-compatibility">Backwards Compatibility<a class="headerlink" href="#backwards-compatibility" title="Permanent link">¶</a></h3>
|
|
<p>Convert from RollingWindowConfig:</p>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-13-1"><a id="__codelineno-13-1" name="__codelineno-13-1" href="#__codelineno-13-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.config</span><span class="w"> </span><span class="kn">import</span> <span class="n">IntelligentContextConfig</span><span class="p">,</span> <span class="n">RollingWindowConfig</span>
|
|
</span><span id="__span-13-2"><a id="__codelineno-13-2" name="__codelineno-13-2" href="#__codelineno-13-2"></a>
|
|
</span><span id="__span-13-3"><a id="__codelineno-13-3" name="__codelineno-13-3" href="#__codelineno-13-3"></a><span class="n">rolling_config</span> <span class="o">=</span> <span class="n">RollingWindowConfig</span><span class="p">(</span>
|
|
</span><span id="__span-13-4"><a id="__codelineno-13-4" name="__codelineno-13-4" href="#__codelineno-13-4"></a> <span class="n">max_tokens</span><span class="o">=</span><span class="mi">100000</span><span class="p">,</span>
|
|
</span><span id="__span-13-5"><a id="__codelineno-13-5" name="__codelineno-13-5" href="#__codelineno-13-5"></a> <span class="n">preserve_system</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span>
|
|
</span><span id="__span-13-6"><a id="__codelineno-13-6" name="__codelineno-13-6" href="#__codelineno-13-6"></a> <span class="n">preserve_recent_turns</span><span class="o">=</span><span class="mi">3</span><span class="p">,</span>
|
|
</span><span id="__span-13-7"><a id="__codelineno-13-7" name="__codelineno-13-7" href="#__codelineno-13-7"></a><span class="p">)</span>
|
|
</span><span id="__span-13-8"><a id="__codelineno-13-8" name="__codelineno-13-8" href="#__codelineno-13-8"></a>
|
|
</span><span id="__span-13-9"><a id="__codelineno-13-9" name="__codelineno-13-9" href="#__codelineno-13-9"></a><span class="c1"># Convert to intelligent context config</span>
|
|
</span><span id="__span-13-10"><a id="__codelineno-13-10" name="__codelineno-13-10" href="#__codelineno-13-10"></a><span class="n">intelligent_config</span> <span class="o">=</span> <span class="n">IntelligentContextConfig</span><span class="p">(</span>
|
|
</span><span id="__span-13-11"><a id="__codelineno-13-11" name="__codelineno-13-11" href="#__codelineno-13-11"></a> <span class="n">keep_system</span><span class="o">=</span><span class="n">rolling_config</span><span class="o">.</span><span class="n">preserve_system</span><span class="p">,</span>
|
|
</span><span id="__span-13-12"><a id="__codelineno-13-12" name="__codelineno-13-12" href="#__codelineno-13-12"></a> <span class="n">keep_last_turns</span><span class="o">=</span><span class="n">rolling_config</span><span class="o">.</span><span class="n">preserve_recent_turns</span><span class="p">,</span>
|
|
</span><span id="__span-13-13"><a id="__codelineno-13-13" name="__codelineno-13-13" href="#__codelineno-13-13"></a><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<hr />
|
|
<h2 id="llmlinguacompressor-optional">LLMLinguaCompressor (Optional)<a class="headerlink" href="#llmlinguacompressor-optional" title="Permanent link">¶</a></h2>
|
|
<p>ML-based compression using Microsoft's LLMLingua-2 model.</p>
|
|
<h3 id="when-to-use">When to Use<a class="headerlink" href="#when-to-use" title="Permanent link">¶</a></h3>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Transform</th>
|
|
<th>Best For</th>
|
|
<th>Speed</th>
|
|
<th>Compression</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>SmartCrusher</td>
|
|
<td>JSON arrays</td>
|
|
<td>~1ms</td>
|
|
<td>70-90%</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Text Utilities</td>
|
|
<td>Search/logs</td>
|
|
<td>~1ms</td>
|
|
<td>50-90%</td>
|
|
</tr>
|
|
<tr>
|
|
<td><strong>LLMLinguaCompressor</strong></td>
|
|
<td>Any text, max compression</td>
|
|
<td>50-200ms</td>
|
|
<td>80-95%</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<h3 id="installation">Installation<a class="headerlink" href="#installation" title="Permanent link">¶</a></h3>
|
|
<div class="language-bash highlight"><pre><span></span><code><span id="__span-14-1"><a id="__codelineno-14-1" name="__codelineno-14-1" href="#__codelineno-14-1"></a>pip<span class="w"> </span>install<span class="w"> </span><span class="s2">"headroom-ai[llmlingua]"</span><span class="w"> </span><span class="c1"># Adds ~2GB</span>
|
|
</span></code></pre></div>
|
|
<h3 id="configuration_4">Configuration<a class="headerlink" href="#configuration_4" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-15-1"><a id="__codelineno-15-1" name="__codelineno-15-1" href="#__codelineno-15-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">LLMLinguaCompressor</span><span class="p">,</span> <span class="n">LLMLinguaConfig</span>
|
|
</span><span id="__span-15-2"><a id="__codelineno-15-2" name="__codelineno-15-2" href="#__codelineno-15-2"></a>
|
|
</span><span id="__span-15-3"><a id="__codelineno-15-3" name="__codelineno-15-3" href="#__codelineno-15-3"></a><span class="n">config</span> <span class="o">=</span> <span class="n">LLMLinguaConfig</span><span class="p">(</span>
|
|
</span><span id="__span-15-4"><a id="__codelineno-15-4" name="__codelineno-15-4" href="#__codelineno-15-4"></a> <span class="n">device</span><span class="o">=</span><span class="s2">"auto"</span><span class="p">,</span> <span class="c1"># auto, cuda, cpu, mps</span>
|
|
</span><span id="__span-15-5"><a id="__codelineno-15-5" name="__codelineno-15-5" href="#__codelineno-15-5"></a> <span class="n">target_compression_rate</span><span class="o">=</span><span class="mf">0.3</span><span class="p">,</span> <span class="c1"># Keep 30% of tokens</span>
|
|
</span><span id="__span-15-6"><a id="__codelineno-15-6" name="__codelineno-15-6" href="#__codelineno-15-6"></a> <span class="n">min_tokens_for_compression</span><span class="o">=</span><span class="mi">100</span><span class="p">,</span> <span class="c1"># Skip small content</span>
|
|
</span><span id="__span-15-7"><a id="__codelineno-15-7" name="__codelineno-15-7" href="#__codelineno-15-7"></a> <span class="n">code_compression_rate</span><span class="o">=</span><span class="mf">0.4</span><span class="p">,</span> <span class="c1"># Conservative for code</span>
|
|
</span><span id="__span-15-8"><a id="__codelineno-15-8" name="__codelineno-15-8" href="#__codelineno-15-8"></a> <span class="n">json_compression_rate</span><span class="o">=</span><span class="mf">0.35</span><span class="p">,</span> <span class="c1"># Moderate for JSON</span>
|
|
</span><span id="__span-15-9"><a id="__codelineno-15-9" name="__codelineno-15-9" href="#__codelineno-15-9"></a> <span class="n">text_compression_rate</span><span class="o">=</span><span class="mf">0.25</span><span class="p">,</span> <span class="c1"># Aggressive for text</span>
|
|
</span><span id="__span-15-10"><a id="__codelineno-15-10" name="__codelineno-15-10" href="#__codelineno-15-10"></a> <span class="n">enable_ccr</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Store original for retrieval</span>
|
|
</span><span id="__span-15-11"><a id="__codelineno-15-11" name="__codelineno-15-11" href="#__codelineno-15-11"></a><span class="p">)</span>
|
|
</span><span id="__span-15-12"><a id="__codelineno-15-12" name="__codelineno-15-12" href="#__codelineno-15-12"></a>
|
|
</span><span id="__span-15-13"><a id="__codelineno-15-13" name="__codelineno-15-13" href="#__codelineno-15-13"></a><span class="n">compressor</span> <span class="o">=</span> <span class="n">LLMLinguaCompressor</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="content-aware-rates">Content-Aware Rates<a class="headerlink" href="#content-aware-rates" title="Permanent link">¶</a></h3>
|
|
<p>LLMLinguaCompressor auto-detects content type:</p>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Content Type</th>
|
|
<th>Default Rate</th>
|
|
<th>Behavior</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>Code</td>
|
|
<td>0.4</td>
|
|
<td>Conservative - preserves syntax</td>
|
|
</tr>
|
|
<tr>
|
|
<td>JSON</td>
|
|
<td>0.35</td>
|
|
<td>Moderate - keeps structure</td>
|
|
</tr>
|
|
<tr>
|
|
<td>Text</td>
|
|
<td>0.3</td>
|
|
<td>Aggressive - maximum compression</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<h3 id="memory-management">Memory Management<a class="headerlink" href="#memory-management" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-16-1"><a id="__codelineno-16-1" name="__codelineno-16-1" href="#__codelineno-16-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span>
|
|
</span><span id="__span-16-2"><a id="__codelineno-16-2" name="__codelineno-16-2" href="#__codelineno-16-2"></a> <span class="n">is_llmlingua_model_loaded</span><span class="p">,</span>
|
|
</span><span id="__span-16-3"><a id="__codelineno-16-3" name="__codelineno-16-3" href="#__codelineno-16-3"></a> <span class="n">unload_llmlingua_model</span><span class="p">,</span>
|
|
</span><span id="__span-16-4"><a id="__codelineno-16-4" name="__codelineno-16-4" href="#__codelineno-16-4"></a><span class="p">)</span>
|
|
</span><span id="__span-16-5"><a id="__codelineno-16-5" name="__codelineno-16-5" href="#__codelineno-16-5"></a>
|
|
</span><span id="__span-16-6"><a id="__codelineno-16-6" name="__codelineno-16-6" href="#__codelineno-16-6"></a><span class="c1"># Check if model is loaded</span>
|
|
</span><span id="__span-16-7"><a id="__codelineno-16-7" name="__codelineno-16-7" href="#__codelineno-16-7"></a><span class="nb">print</span><span class="p">(</span><span class="n">is_llmlingua_model_loaded</span><span class="p">())</span> <span class="c1"># True/False</span>
|
|
</span><span id="__span-16-8"><a id="__codelineno-16-8" name="__codelineno-16-8" href="#__codelineno-16-8"></a>
|
|
</span><span id="__span-16-9"><a id="__codelineno-16-9" name="__codelineno-16-9" href="#__codelineno-16-9"></a><span class="c1"># Free ~1GB RAM when done</span>
|
|
</span><span id="__span-16-10"><a id="__codelineno-16-10" name="__codelineno-16-10" href="#__codelineno-16-10"></a><span class="n">unload_llmlingua_model</span><span class="p">()</span>
|
|
</span></code></pre></div>
|
|
<h3 id="proxy-integration">Proxy Integration<a class="headerlink" href="#proxy-integration" title="Permanent link">¶</a></h3>
|
|
<div class="language-bash highlight"><pre><span></span><code><span id="__span-17-1"><a id="__codelineno-17-1" name="__codelineno-17-1" href="#__codelineno-17-1"></a><span class="c1"># Enable in proxy</span>
|
|
</span><span id="__span-17-2"><a id="__codelineno-17-2" name="__codelineno-17-2" href="#__codelineno-17-2"></a>headroom<span class="w"> </span>proxy<span class="w"> </span>--llmlingua<span class="w"> </span>--llmlingua-device<span class="w"> </span>cuda<span class="w"> </span>--llmlingua-rate<span class="w"> </span><span class="m">0</span>.3
|
|
</span></code></pre></div>
|
|
<hr />
|
|
<h2 id="codeawarecompressor-optional">CodeAwareCompressor (Optional)<a class="headerlink" href="#codeawarecompressor-optional" title="Permanent link">¶</a></h2>
|
|
<p>AST-based compression for source code using tree-sitter.</p>
|
|
<h3 id="when-to-use_1">When to Use<a class="headerlink" href="#when-to-use_1" title="Permanent link">¶</a></h3>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Transform</th>
|
|
<th>Best For</th>
|
|
<th>Speed</th>
|
|
<th>Compression</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>SmartCrusher</td>
|
|
<td>JSON arrays</td>
|
|
<td>~1ms</td>
|
|
<td>70-90%</td>
|
|
</tr>
|
|
<tr>
|
|
<td><strong>CodeAwareCompressor</strong></td>
|
|
<td>Source code</td>
|
|
<td>~10-50ms</td>
|
|
<td>40-70%</td>
|
|
</tr>
|
|
<tr>
|
|
<td>LLMLinguaCompressor</td>
|
|
<td>Any text</td>
|
|
<td>50-200ms</td>
|
|
<td>80-95%</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<h3 id="key-benefits">Key Benefits<a class="headerlink" href="#key-benefits" title="Permanent link">¶</a></h3>
|
|
<ul>
|
|
<li><strong>Syntax validity guaranteed</strong> — Output always parses correctly</li>
|
|
<li><strong>Preserves critical structure</strong> — Imports, signatures, types, error handlers</li>
|
|
<li><strong>Multi-language support</strong> — Python, JavaScript, TypeScript, Go, Rust, Java, C, C++</li>
|
|
<li><strong>Lightweight</strong> — ~50MB vs ~1GB for LLMLingua</li>
|
|
</ul>
|
|
<h3 id="installation_1">Installation<a class="headerlink" href="#installation_1" title="Permanent link">¶</a></h3>
|
|
<div class="language-bash highlight"><pre><span></span><code><span id="__span-18-1"><a id="__codelineno-18-1" name="__codelineno-18-1" href="#__codelineno-18-1"></a>pip<span class="w"> </span>install<span class="w"> </span><span class="s2">"headroom-ai[code]"</span><span class="w"> </span><span class="c1"># Adds tree-sitter-language-pack</span>
|
|
</span></code></pre></div>
|
|
<h3 id="configuration_5">Configuration<a class="headerlink" href="#configuration_5" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-19-1"><a id="__codelineno-19-1" name="__codelineno-19-1" href="#__codelineno-19-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">CodeAwareCompressor</span><span class="p">,</span> <span class="n">CodeCompressorConfig</span><span class="p">,</span> <span class="n">DocstringMode</span>
|
|
</span><span id="__span-19-2"><a id="__codelineno-19-2" name="__codelineno-19-2" href="#__codelineno-19-2"></a>
|
|
</span><span id="__span-19-3"><a id="__codelineno-19-3" name="__codelineno-19-3" href="#__codelineno-19-3"></a><span class="n">config</span> <span class="o">=</span> <span class="n">CodeCompressorConfig</span><span class="p">(</span>
|
|
</span><span id="__span-19-4"><a id="__codelineno-19-4" name="__codelineno-19-4" href="#__codelineno-19-4"></a> <span class="n">preserve_imports</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Always keep imports</span>
|
|
</span><span id="__span-19-5"><a id="__codelineno-19-5" name="__codelineno-19-5" href="#__codelineno-19-5"></a> <span class="n">preserve_signatures</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Always keep function signatures</span>
|
|
</span><span id="__span-19-6"><a id="__codelineno-19-6" name="__codelineno-19-6" href="#__codelineno-19-6"></a> <span class="n">preserve_type_annotations</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Keep type hints</span>
|
|
</span><span id="__span-19-7"><a id="__codelineno-19-7" name="__codelineno-19-7" href="#__codelineno-19-7"></a> <span class="n">preserve_error_handlers</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Keep try/except blocks</span>
|
|
</span><span id="__span-19-8"><a id="__codelineno-19-8" name="__codelineno-19-8" href="#__codelineno-19-8"></a> <span class="n">preserve_decorators</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Keep decorators</span>
|
|
</span><span id="__span-19-9"><a id="__codelineno-19-9" name="__codelineno-19-9" href="#__codelineno-19-9"></a> <span class="n">docstring_mode</span><span class="o">=</span><span class="n">DocstringMode</span><span class="o">.</span><span class="n">FIRST_LINE</span><span class="p">,</span> <span class="c1"># FULL, FIRST_LINE, REMOVE</span>
|
|
</span><span id="__span-19-10"><a id="__codelineno-19-10" name="__codelineno-19-10" href="#__codelineno-19-10"></a> <span class="n">target_compression_rate</span><span class="o">=</span><span class="mf">0.2</span><span class="p">,</span> <span class="c1"># Keep 20% of tokens</span>
|
|
</span><span id="__span-19-11"><a id="__codelineno-19-11" name="__codelineno-19-11" href="#__codelineno-19-11"></a> <span class="n">max_body_lines</span><span class="o">=</span><span class="mi">5</span><span class="p">,</span> <span class="c1"># Lines to keep per function body</span>
|
|
</span><span id="__span-19-12"><a id="__codelineno-19-12" name="__codelineno-19-12" href="#__codelineno-19-12"></a> <span class="n">min_tokens_for_compression</span><span class="o">=</span><span class="mi">100</span><span class="p">,</span> <span class="c1"># Skip small content</span>
|
|
</span><span id="__span-19-13"><a id="__codelineno-19-13" name="__codelineno-19-13" href="#__codelineno-19-13"></a> <span class="n">language_hint</span><span class="o">=</span><span class="kc">None</span><span class="p">,</span> <span class="c1"># Auto-detect if None</span>
|
|
</span><span id="__span-19-14"><a id="__codelineno-19-14" name="__codelineno-19-14" href="#__codelineno-19-14"></a> <span class="n">fallback_to_llmlingua</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Use LLMLingua for unknown langs</span>
|
|
</span><span id="__span-19-15"><a id="__codelineno-19-15" name="__codelineno-19-15" href="#__codelineno-19-15"></a><span class="p">)</span>
|
|
</span><span id="__span-19-16"><a id="__codelineno-19-16" name="__codelineno-19-16" href="#__codelineno-19-16"></a>
|
|
</span><span id="__span-19-17"><a id="__codelineno-19-17" name="__codelineno-19-17" href="#__codelineno-19-17"></a><span class="n">compressor</span> <span class="o">=</span> <span class="n">CodeAwareCompressor</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="example_1">Example<a class="headerlink" href="#example_1" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-20-1"><a id="__codelineno-20-1" name="__codelineno-20-1" href="#__codelineno-20-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">CodeAwareCompressor</span>
|
|
</span><span id="__span-20-2"><a id="__codelineno-20-2" name="__codelineno-20-2" href="#__codelineno-20-2"></a>
|
|
</span><span id="__span-20-3"><a id="__codelineno-20-3" name="__codelineno-20-3" href="#__codelineno-20-3"></a><span class="n">compressor</span> <span class="o">=</span> <span class="n">CodeAwareCompressor</span><span class="p">()</span>
|
|
</span><span id="__span-20-4"><a id="__codelineno-20-4" name="__codelineno-20-4" href="#__codelineno-20-4"></a>
|
|
</span><span id="__span-20-5"><a id="__codelineno-20-5" name="__codelineno-20-5" href="#__codelineno-20-5"></a><span class="n">code</span> <span class="o">=</span> <span class="s1">'''</span>
|
|
</span><span id="__span-20-6"><a id="__codelineno-20-6" name="__codelineno-20-6" href="#__codelineno-20-6"></a><span class="s1">import os</span>
|
|
</span><span id="__span-20-7"><a id="__codelineno-20-7" name="__codelineno-20-7" href="#__codelineno-20-7"></a><span class="s1">from typing import List</span>
|
|
</span><span id="__span-20-8"><a id="__codelineno-20-8" name="__codelineno-20-8" href="#__codelineno-20-8"></a>
|
|
</span><span id="__span-20-9"><a id="__codelineno-20-9" name="__codelineno-20-9" href="#__codelineno-20-9"></a><span class="s1">def process_items(items: List[str]) -> List[str]:</span>
|
|
</span><span id="__span-20-10"><a id="__codelineno-20-10" name="__codelineno-20-10" href="#__codelineno-20-10"></a><span class="s1"> """Process a list of items."""</span>
|
|
</span><span id="__span-20-11"><a id="__codelineno-20-11" name="__codelineno-20-11" href="#__codelineno-20-11"></a><span class="s1"> results = []</span>
|
|
</span><span id="__span-20-12"><a id="__codelineno-20-12" name="__codelineno-20-12" href="#__codelineno-20-12"></a><span class="s1"> for item in items:</span>
|
|
</span><span id="__span-20-13"><a id="__codelineno-20-13" name="__codelineno-20-13" href="#__codelineno-20-13"></a><span class="s1"> if not item:</span>
|
|
</span><span id="__span-20-14"><a id="__codelineno-20-14" name="__codelineno-20-14" href="#__codelineno-20-14"></a><span class="s1"> continue</span>
|
|
</span><span id="__span-20-15"><a id="__codelineno-20-15" name="__codelineno-20-15" href="#__codelineno-20-15"></a><span class="s1"> processed = item.strip().lower()</span>
|
|
</span><span id="__span-20-16"><a id="__codelineno-20-16" name="__codelineno-20-16" href="#__codelineno-20-16"></a><span class="s1"> results.append(processed)</span>
|
|
</span><span id="__span-20-17"><a id="__codelineno-20-17" name="__codelineno-20-17" href="#__codelineno-20-17"></a><span class="s1"> return results</span>
|
|
</span><span id="__span-20-18"><a id="__codelineno-20-18" name="__codelineno-20-18" href="#__codelineno-20-18"></a><span class="s1">'''</span>
|
|
</span><span id="__span-20-19"><a id="__codelineno-20-19" name="__codelineno-20-19" href="#__codelineno-20-19"></a>
|
|
</span><span id="__span-20-20"><a id="__codelineno-20-20" name="__codelineno-20-20" href="#__codelineno-20-20"></a><span class="n">result</span> <span class="o">=</span> <span class="n">compressor</span><span class="o">.</span><span class="n">compress</span><span class="p">(</span><span class="n">code</span><span class="p">,</span> <span class="n">language</span><span class="o">=</span><span class="s2">"python"</span><span class="p">)</span>
|
|
</span><span id="__span-20-21"><a id="__codelineno-20-21" name="__codelineno-20-21" href="#__codelineno-20-21"></a><span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="o">.</span><span class="n">compressed</span><span class="p">)</span>
|
|
</span><span id="__span-20-22"><a id="__codelineno-20-22" name="__codelineno-20-22" href="#__codelineno-20-22"></a><span class="c1"># import os</span>
|
|
</span><span id="__span-20-23"><a id="__codelineno-20-23" name="__codelineno-20-23" href="#__codelineno-20-23"></a><span class="c1"># from typing import List</span>
|
|
</span><span id="__span-20-24"><a id="__codelineno-20-24" name="__codelineno-20-24" href="#__codelineno-20-24"></a><span class="c1">#</span>
|
|
</span><span id="__span-20-25"><a id="__codelineno-20-25" name="__codelineno-20-25" href="#__codelineno-20-25"></a><span class="c1"># def process_items(items: List[str]) -> List[str]:</span>
|
|
</span><span id="__span-20-26"><a id="__codelineno-20-26" name="__codelineno-20-26" href="#__codelineno-20-26"></a><span class="c1"># """Process a list of items."""</span>
|
|
</span><span id="__span-20-27"><a id="__codelineno-20-27" name="__codelineno-20-27" href="#__codelineno-20-27"></a><span class="c1"># results = []</span>
|
|
</span><span id="__span-20-28"><a id="__codelineno-20-28" name="__codelineno-20-28" href="#__codelineno-20-28"></a><span class="c1"># for item in items:</span>
|
|
</span><span id="__span-20-29"><a id="__codelineno-20-29" name="__codelineno-20-29" href="#__codelineno-20-29"></a><span class="c1"># # ... (5 lines compressed)</span>
|
|
</span><span id="__span-20-30"><a id="__codelineno-20-30" name="__codelineno-20-30" href="#__codelineno-20-30"></a><span class="c1"># pass</span>
|
|
</span><span id="__span-20-31"><a id="__codelineno-20-31" name="__codelineno-20-31" href="#__codelineno-20-31"></a>
|
|
</span><span id="__span-20-32"><a id="__codelineno-20-32" name="__codelineno-20-32" href="#__codelineno-20-32"></a><span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Compression: </span><span class="si">{</span><span class="n">result</span><span class="o">.</span><span class="n">compression_ratio</span><span class="si">:</span><span class="s2">.0%</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span> <span class="c1"># ~55%</span>
|
|
</span><span id="__span-20-33"><a id="__codelineno-20-33" name="__codelineno-20-33" href="#__codelineno-20-33"></a><span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Syntax valid: </span><span class="si">{</span><span class="n">result</span><span class="o">.</span><span class="n">syntax_valid</span><span class="si">}</span><span class="s2">"</span><span class="p">)</span> <span class="c1"># True</span>
|
|
</span></code></pre></div>
|
|
<h3 id="supported-languages">Supported Languages<a class="headerlink" href="#supported-languages" title="Permanent link">¶</a></h3>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Tier</th>
|
|
<th>Languages</th>
|
|
<th>Support Level</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>1</td>
|
|
<td>Python, JavaScript, TypeScript</td>
|
|
<td>Full AST analysis</td>
|
|
</tr>
|
|
<tr>
|
|
<td>2</td>
|
|
<td>Go, Rust, Java, C, C++</td>
|
|
<td>Function body compression</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<h3 id="memory-management_1">Memory Management<a class="headerlink" href="#memory-management_1" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-21-1"><a id="__codelineno-21-1" name="__codelineno-21-1" href="#__codelineno-21-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">is_tree_sitter_available</span><span class="p">,</span> <span class="n">unload_tree_sitter</span>
|
|
</span><span id="__span-21-2"><a id="__codelineno-21-2" name="__codelineno-21-2" href="#__codelineno-21-2"></a>
|
|
</span><span id="__span-21-3"><a id="__codelineno-21-3" name="__codelineno-21-3" href="#__codelineno-21-3"></a><span class="c1"># Check if tree-sitter is installed</span>
|
|
</span><span id="__span-21-4"><a id="__codelineno-21-4" name="__codelineno-21-4" href="#__codelineno-21-4"></a><span class="nb">print</span><span class="p">(</span><span class="n">is_tree_sitter_available</span><span class="p">())</span> <span class="c1"># True/False</span>
|
|
</span><span id="__span-21-5"><a id="__codelineno-21-5" name="__codelineno-21-5" href="#__codelineno-21-5"></a>
|
|
</span><span id="__span-21-6"><a id="__codelineno-21-6" name="__codelineno-21-6" href="#__codelineno-21-6"></a><span class="c1"># Free memory when done (parsers are lazy-loaded)</span>
|
|
</span><span id="__span-21-7"><a id="__codelineno-21-7" name="__codelineno-21-7" href="#__codelineno-21-7"></a><span class="n">unload_tree_sitter</span><span class="p">()</span>
|
|
</span></code></pre></div>
|
|
<hr />
|
|
<h2 id="contentrouter">ContentRouter<a class="headerlink" href="#contentrouter" title="Permanent link">¶</a></h2>
|
|
<p>Intelligent compression orchestrator that routes content to the optimal compressor.</p>
|
|
<h3 id="how-it-works_1">How It Works<a class="headerlink" href="#how-it-works_1" title="Permanent link">¶</a></h3>
|
|
<p>ContentRouter analyzes content and selects the best compression strategy:</p>
|
|
<ol>
|
|
<li><strong>Detect content type</strong> — JSON, code, logs, search results, plain text</li>
|
|
<li><strong>Consider source hints</strong> — File paths, tool names for high-confidence routing</li>
|
|
<li><strong>Route to compressor</strong> — SmartCrusher, CodeAwareCompressor, SearchCompressor, etc.</li>
|
|
<li><strong>Log decisions</strong> — Transparent routing for debugging</li>
|
|
</ol>
|
|
<h3 id="configuration_6">Configuration<a class="headerlink" href="#configuration_6" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-22-1"><a id="__codelineno-22-1" name="__codelineno-22-1" href="#__codelineno-22-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">ContentRouter</span><span class="p">,</span> <span class="n">ContentRouterConfig</span><span class="p">,</span> <span class="n">CompressionStrategy</span>
|
|
</span><span id="__span-22-2"><a id="__codelineno-22-2" name="__codelineno-22-2" href="#__codelineno-22-2"></a>
|
|
</span><span id="__span-22-3"><a id="__codelineno-22-3" name="__codelineno-22-3" href="#__codelineno-22-3"></a><span class="n">config</span> <span class="o">=</span> <span class="n">ContentRouterConfig</span><span class="p">(</span>
|
|
</span><span id="__span-22-4"><a id="__codelineno-22-4" name="__codelineno-22-4" href="#__codelineno-22-4"></a> <span class="n">min_section_tokens</span><span class="o">=</span><span class="mi">100</span><span class="p">,</span> <span class="c1"># Minimum tokens to compress</span>
|
|
</span><span id="__span-22-5"><a id="__codelineno-22-5" name="__codelineno-22-5" href="#__codelineno-22-5"></a> <span class="n">enable_code_aware</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Use CodeAwareCompressor for code</span>
|
|
</span><span id="__span-22-6"><a id="__codelineno-22-6" name="__codelineno-22-6" href="#__codelineno-22-6"></a> <span class="n">enable_search_compression</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Use SearchCompressor for grep output</span>
|
|
</span><span id="__span-22-7"><a id="__codelineno-22-7" name="__codelineno-22-7" href="#__codelineno-22-7"></a> <span class="n">enable_log_compression</span><span class="o">=</span><span class="kc">True</span><span class="p">,</span> <span class="c1"># Use LogCompressor for logs</span>
|
|
</span><span id="__span-22-8"><a id="__codelineno-22-8" name="__codelineno-22-8" href="#__codelineno-22-8"></a> <span class="n">default_strategy</span><span class="o">=</span><span class="n">CompressionStrategy</span><span class="o">.</span><span class="n">TEXT</span><span class="p">,</span> <span class="c1"># Fallback strategy</span>
|
|
</span><span id="__span-22-9"><a id="__codelineno-22-9" name="__codelineno-22-9" href="#__codelineno-22-9"></a><span class="p">)</span>
|
|
</span><span id="__span-22-10"><a id="__codelineno-22-10" name="__codelineno-22-10" href="#__codelineno-22-10"></a>
|
|
</span><span id="__span-22-11"><a id="__codelineno-22-11" name="__codelineno-22-11" href="#__codelineno-22-11"></a><span class="n">router</span> <span class="o">=</span> <span class="n">ContentRouter</span><span class="p">(</span><span class="n">config</span><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="example_2">Example<a class="headerlink" href="#example_2" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-23-1"><a id="__codelineno-23-1" name="__codelineno-23-1" href="#__codelineno-23-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="n">ContentRouter</span>
|
|
</span><span id="__span-23-2"><a id="__codelineno-23-2" name="__codelineno-23-2" href="#__codelineno-23-2"></a>
|
|
</span><span id="__span-23-3"><a id="__codelineno-23-3" name="__codelineno-23-3" href="#__codelineno-23-3"></a><span class="n">router</span> <span class="o">=</span> <span class="n">ContentRouter</span><span class="p">()</span>
|
|
</span><span id="__span-23-4"><a id="__codelineno-23-4" name="__codelineno-23-4" href="#__codelineno-23-4"></a>
|
|
</span><span id="__span-23-5"><a id="__codelineno-23-5" name="__codelineno-23-5" href="#__codelineno-23-5"></a><span class="c1"># Router auto-detects content type and routes to optimal compressor</span>
|
|
</span><span id="__span-23-6"><a id="__codelineno-23-6" name="__codelineno-23-6" href="#__codelineno-23-6"></a><span class="n">result</span> <span class="o">=</span> <span class="n">router</span><span class="o">.</span><span class="n">compress</span><span class="p">(</span><span class="n">content</span><span class="p">)</span>
|
|
</span><span id="__span-23-7"><a id="__codelineno-23-7" name="__codelineno-23-7" href="#__codelineno-23-7"></a>
|
|
</span><span id="__span-23-8"><a id="__codelineno-23-8" name="__codelineno-23-8" href="#__codelineno-23-8"></a><span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="o">.</span><span class="n">strategy_used</span><span class="p">)</span> <span class="c1"># CompressionStrategy.CODE_AWARE, SMART_CRUSHER, etc.</span>
|
|
</span><span id="__span-23-9"><a id="__codelineno-23-9" name="__codelineno-23-9" href="#__codelineno-23-9"></a><span class="nb">print</span><span class="p">(</span><span class="n">result</span><span class="o">.</span><span class="n">routing_log</span><span class="p">)</span> <span class="c1"># List of routing decisions</span>
|
|
</span></code></pre></div>
|
|
<h3 id="compression-strategies">Compression Strategies<a class="headerlink" href="#compression-strategies" title="Permanent link">¶</a></h3>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Strategy</th>
|
|
<th>Used For</th>
|
|
<th>Compressor</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>CODE_AWARE</td>
|
|
<td>Source code</td>
|
|
<td>CodeAwareCompressor</td>
|
|
</tr>
|
|
<tr>
|
|
<td>SMART_CRUSHER</td>
|
|
<td>JSON arrays</td>
|
|
<td>SmartCrusher</td>
|
|
</tr>
|
|
<tr>
|
|
<td>SEARCH</td>
|
|
<td>Grep/find output</td>
|
|
<td>SearchCompressor</td>
|
|
</tr>
|
|
<tr>
|
|
<td>LOG</td>
|
|
<td>Log files</td>
|
|
<td>LogCompressor</td>
|
|
</tr>
|
|
<tr>
|
|
<td>TEXT</td>
|
|
<td>Plain text</td>
|
|
<td>TextCompressor</td>
|
|
</tr>
|
|
<tr>
|
|
<td>LLMLINGUA</td>
|
|
<td>Any (max compression)</td>
|
|
<td>LLMLinguaCompressor</td>
|
|
</tr>
|
|
<tr>
|
|
<td>PASSTHROUGH</td>
|
|
<td>Small content</td>
|
|
<td>None</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<h3 id="content-detection">Content Detection<a class="headerlink" href="#content-detection" title="Permanent link">¶</a></h3>
|
|
<p>The router automatically detects content types by analyzing the content itself:</p>
|
|
<ul>
|
|
<li><strong>Source code</strong>: Detected by syntax patterns, indentation, keywords</li>
|
|
<li><strong>JSON arrays</strong>: Detected by JSON structure with array elements</li>
|
|
<li><strong>Search results</strong>: Detected by <code>file:line:</code> patterns</li>
|
|
<li><strong>Log output</strong>: Detected by timestamp and log level patterns</li>
|
|
<li><strong>Plain text</strong>: Fallback for prose content</li>
|
|
</ul>
|
|
<p>No manual hints required - the router inspects content directly.</p>
|
|
<h3 id="toin-integration">TOIN Integration<a class="headerlink" href="#toin-integration" title="Permanent link">¶</a></h3>
|
|
<p>ContentRouter records all compressions to TOIN (Tool Output Intelligence Network) for cross-user learning:</p>
|
|
<ul>
|
|
<li><strong>All strategies tracked</strong>: Code, search, logs, text, and LLMLingua compressions are recorded</li>
|
|
<li><strong>Retrieval feedback</strong>: When users retrieve original content via CCR, TOIN learns which compressions need expansion</li>
|
|
<li><strong>Pattern learning</strong>: TOIN builds signatures for each content type to improve future compressions</li>
|
|
</ul>
|
|
<p>This enables the feedback loop where compression decisions improve based on actual user behavior across all content types, not just JSON arrays.</p>
|
|
<hr />
|
|
<h2 id="transformpipeline">TransformPipeline<a class="headerlink" href="#transformpipeline" title="Permanent link">¶</a></h2>
|
|
<p>Combine transforms for optimal results.</p>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-24-1"><a id="__codelineno-24-1" name="__codelineno-24-1" href="#__codelineno-24-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom</span><span class="w"> </span><span class="kn">import</span> <span class="n">TransformPipeline</span><span class="p">,</span> <span class="n">SmartCrusher</span><span class="p">,</span> <span class="n">CacheAligner</span><span class="p">,</span> <span class="n">RollingWindow</span>
|
|
</span><span id="__span-24-2"><a id="__codelineno-24-2" name="__codelineno-24-2" href="#__codelineno-24-2"></a>
|
|
</span><span id="__span-24-3"><a id="__codelineno-24-3" name="__codelineno-24-3" href="#__codelineno-24-3"></a><span class="n">pipeline</span> <span class="o">=</span> <span class="n">TransformPipeline</span><span class="p">([</span>
|
|
</span><span id="__span-24-4"><a id="__codelineno-24-4" name="__codelineno-24-4" href="#__codelineno-24-4"></a> <span class="n">SmartCrusher</span><span class="p">(),</span> <span class="c1"># First: compress tool outputs</span>
|
|
</span><span id="__span-24-5"><a id="__codelineno-24-5" name="__codelineno-24-5" href="#__codelineno-24-5"></a> <span class="n">CacheAligner</span><span class="p">(),</span> <span class="c1"># Then: stabilize prefix</span>
|
|
</span><span id="__span-24-6"><a id="__codelineno-24-6" name="__codelineno-24-6" href="#__codelineno-24-6"></a> <span class="n">RollingWindow</span><span class="p">(),</span> <span class="c1"># Finally: fit in context</span>
|
|
</span><span id="__span-24-7"><a id="__codelineno-24-7" name="__codelineno-24-7" href="#__codelineno-24-7"></a><span class="p">])</span>
|
|
</span><span id="__span-24-8"><a id="__codelineno-24-8" name="__codelineno-24-8" href="#__codelineno-24-8"></a>
|
|
</span><span id="__span-24-9"><a id="__codelineno-24-9" name="__codelineno-24-9" href="#__codelineno-24-9"></a><span class="n">result</span> <span class="o">=</span> <span class="n">pipeline</span><span class="o">.</span><span class="n">transform</span><span class="p">(</span><span class="n">messages</span><span class="p">)</span>
|
|
</span><span id="__span-24-10"><a id="__codelineno-24-10" name="__codelineno-24-10" href="#__codelineno-24-10"></a><span class="nb">print</span><span class="p">(</span><span class="sa">f</span><span class="s2">"Saved </span><span class="si">{</span><span class="n">result</span><span class="o">.</span><span class="n">tokens_saved</span><span class="si">}</span><span class="s2"> tokens"</span><span class="p">)</span>
|
|
</span></code></pre></div>
|
|
<h3 id="with-llmlingua-optional">With LLMLingua (Optional)<a class="headerlink" href="#with-llmlingua-optional" title="Permanent link">¶</a></h3>
|
|
<div class="language-python highlight"><pre><span></span><code><span id="__span-25-1"><a id="__codelineno-25-1" name="__codelineno-25-1" href="#__codelineno-25-1"></a><span class="kn">from</span><span class="w"> </span><span class="nn">headroom.transforms</span><span class="w"> </span><span class="kn">import</span> <span class="p">(</span>
|
|
</span><span id="__span-25-2"><a id="__codelineno-25-2" name="__codelineno-25-2" href="#__codelineno-25-2"></a> <span class="n">TransformPipeline</span><span class="p">,</span> <span class="n">SmartCrusher</span><span class="p">,</span> <span class="n">CacheAligner</span><span class="p">,</span>
|
|
</span><span id="__span-25-3"><a id="__codelineno-25-3" name="__codelineno-25-3" href="#__codelineno-25-3"></a> <span class="n">RollingWindow</span><span class="p">,</span> <span class="n">LLMLinguaCompressor</span>
|
|
</span><span id="__span-25-4"><a id="__codelineno-25-4" name="__codelineno-25-4" href="#__codelineno-25-4"></a><span class="p">)</span>
|
|
</span><span id="__span-25-5"><a id="__codelineno-25-5" name="__codelineno-25-5" href="#__codelineno-25-5"></a>
|
|
</span><span id="__span-25-6"><a id="__codelineno-25-6" name="__codelineno-25-6" href="#__codelineno-25-6"></a><span class="n">pipeline</span> <span class="o">=</span> <span class="n">TransformPipeline</span><span class="p">([</span>
|
|
</span><span id="__span-25-7"><a id="__codelineno-25-7" name="__codelineno-25-7" href="#__codelineno-25-7"></a> <span class="n">CacheAligner</span><span class="p">(),</span> <span class="c1"># 1. Stabilize prefix</span>
|
|
</span><span id="__span-25-8"><a id="__codelineno-25-8" name="__codelineno-25-8" href="#__codelineno-25-8"></a> <span class="n">SmartCrusher</span><span class="p">(),</span> <span class="c1"># 2. Compress JSON arrays</span>
|
|
</span><span id="__span-25-9"><a id="__codelineno-25-9" name="__codelineno-25-9" href="#__codelineno-25-9"></a> <span class="n">LLMLinguaCompressor</span><span class="p">(),</span> <span class="c1"># 3. ML compression on remaining text</span>
|
|
</span><span id="__span-25-10"><a id="__codelineno-25-10" name="__codelineno-25-10" href="#__codelineno-25-10"></a> <span class="n">RollingWindow</span><span class="p">(),</span> <span class="c1"># 4. Final size constraint (always last)</span>
|
|
</span><span id="__span-25-11"><a id="__codelineno-25-11" name="__codelineno-25-11" href="#__codelineno-25-11"></a><span class="p">])</span>
|
|
</span></code></pre></div>
|
|
<h3 id="recommended-order">Recommended Order<a class="headerlink" href="#recommended-order" title="Permanent link">¶</a></h3>
|
|
<table>
|
|
<thead>
|
|
<tr>
|
|
<th>Order</th>
|
|
<th>Transform</th>
|
|
<th>Purpose</th>
|
|
</tr>
|
|
</thead>
|
|
<tbody>
|
|
<tr>
|
|
<td>1</td>
|
|
<td>CacheAligner</td>
|
|
<td>Stabilize prefix for caching</td>
|
|
</tr>
|
|
<tr>
|
|
<td>2</td>
|
|
<td>SmartCrusher</td>
|
|
<td>Compress JSON tool outputs</td>
|
|
</tr>
|
|
<tr>
|
|
<td>3</td>
|
|
<td>LLMLinguaCompressor</td>
|
|
<td>ML compression (optional)</td>
|
|
</tr>
|
|
<tr>
|
|
<td>4</td>
|
|
<td>RollingWindow</td>
|
|
<td>Enforce token limits (always last)</td>
|
|
</tr>
|
|
</tbody>
|
|
</table>
|
|
<p><strong>Why this order?</strong>
|
|
- CacheAligner first to maximize prefix stability
|
|
- SmartCrusher handles JSON arrays efficiently
|
|
- LLMLingua compresses remaining long text
|
|
- RollingWindow truncates only if still over limit</p>
|
|
<hr />
|
|
<h2 id="safety-guarantees">Safety Guarantees<a class="headerlink" href="#safety-guarantees" title="Permanent link">¶</a></h2>
|
|
<p>All transforms follow strict safety rules:</p>
|
|
<ol>
|
|
<li><strong>Never remove human content</strong> - User/assistant text is sacred</li>
|
|
<li><strong>Never break tool ordering</strong> - Calls and results stay paired</li>
|
|
<li><strong>Parse failures are no-ops</strong> - Malformed content passes through</li>
|
|
<li><strong>Preserves recency</strong> - Last N turns always kept</li>
|
|
<li><strong>100% error preservation</strong> - Error items never dropped</li>
|
|
</ol>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
</article>
|
|
</div>
|
|
|
|
|
|
<script>var tabs=__md_get("__tabs");if(Array.isArray(tabs))e:for(var set of document.querySelectorAll(".tabbed-set")){var labels=set.querySelector(".tabbed-labels");for(var tab of tabs)for(var label of labels.getElementsByTagName("label"))if(label.innerText.trim()===tab){var input=document.getElementById(label.htmlFor);input.checked=!0;continue e}}</script>
|
|
|
|
<script>var target=document.getElementById(location.hash.slice(1));target&&target.name&&(target.checked=target.name.startsWith("__tabbed_"))</script>
|
|
</div>
|
|
|
|
<button type="button" class="md-top md-icon" data-md-component="top" hidden>
|
|
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="M13 20h-2V8l-5.5 5.5-1.42-1.42L12 4.16l7.92 7.92-1.42 1.42L13 8z"/></svg>
|
|
Back to top
|
|
</button>
|
|
|
|
</main>
|
|
|
|
<footer class="md-footer">
|
|
|
|
<div class="md-footer-meta md-typeset">
|
|
<div class="md-footer-meta__inner md-grid">
|
|
<div class="md-copyright">
|
|
|
|
|
|
Made with
|
|
<a href="https://squidfunk.github.io/mkdocs-material/" target="_blank" rel="noopener">
|
|
Material for MkDocs
|
|
</a>
|
|
|
|
</div>
|
|
|
|
|
|
<div class="md-social">
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<a href="https://github.com/chopratejas/headroom" target="_blank" rel="noopener" title="github.com" class="md-social__link">
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 512 512"><!--! Font Awesome Free 7.1.0 by @fontawesome - https://fontawesome.com License - https://fontawesome.com/license/free (Icons: CC BY 4.0, Fonts: SIL OFL 1.1, Code: MIT License) Copyright 2025 Fonticons, Inc.--><path d="M173.9 397.4c0 2-2.3 3.6-5.2 3.6-3.3.3-5.6-1.3-5.6-3.6 0-2 2.3-3.6 5.2-3.6 3-.3 5.6 1.3 5.6 3.6m-31.1-4.5c-.7 2 1.3 4.3 4.3 4.9 2.6 1 5.6 0 6.2-2s-1.3-4.3-4.3-5.2c-2.6-.7-5.5.3-6.2 2.3m44.2-1.7c-2.9.7-4.9 2.6-4.6 4.9.3 2 2.9 3.3 5.9 2.6 2.9-.7 4.9-2.6 4.6-4.6-.3-1.9-3-3.2-5.9-2.9M252.8 8C114.1 8 8 113.3 8 252c0 110.9 69.8 205.8 169.5 239.2 12.8 2.3 17.3-5.6 17.3-12.1 0-6.2-.3-40.4-.3-61.4 0 0-70 15-84.7-29.8 0 0-11.4-29.1-27.8-36.6 0 0-22.9-15.7 1.6-15.4 0 0 24.9 2 38.6 25.8 21.9 38.6 58.6 27.5 72.9 20.9 2.3-16 8.8-27.1 16-33.7-55.9-6.2-112.3-14.3-112.3-110.5 0-27.5 7.6-41.3 23.6-58.9-2.6-6.5-11.1-33.3 2.6-67.9 20.9-6.5 69 27 69 27 20-5.6 41.5-8.5 62.8-8.5s42.8 2.9 62.8 8.5c0 0 48.1-33.6 69-27 13.7 34.7 5.2 61.4 2.6 67.9 16 17.7 25.8 31.5 25.8 58.9 0 96.5-58.9 104.2-114.8 110.5 9.2 7.9 17 22.9 17 46.4 0 33.7-.3 75.4-.3 83.6 0 6.5 4.6 14.4 17.3 12.1C436.2 457.8 504 362.9 504 252 504 113.3 391.5 8 252.8 8M105.2 352.9c-1.3 1-1 3.3.7 5.2 1.6 1.6 3.9 2.3 5.2 1 1.3-1 1-3.3-.7-5.2-1.6-1.6-3.9-2.3-5.2-1m-10.8-8.1c-.7 1.3.3 2.9 2.3 3.9 1.6 1 3.6.7 4.3-.7.7-1.3-.3-2.9-2.3-3.9-2-.6-3.6-.3-4.3.7m32.4 35.6c-1.6 1.3-1 4.3 1.3 6.2 2.3 2.3 5.2 2.6 6.5 1 1.3-1.3.7-4.3-1.3-6.2-2.2-2.3-5.2-2.6-6.5-1m-11.4-14.7c-1.6 1-1.6 3.6 0 5.9s4.3 3.3 5.6 2.3c1.6-1.3 1.6-3.9 0-6.2-1.4-2.3-4-3.3-5.6-2"/></svg>
|
|
</a>
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
<a href="https://pypi.org/project/headroom-ai/" target="_blank" rel="noopener" title="pypi.org" class="md-social__link">
|
|
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 448 512"><!--! Font Awesome Free 7.1.0 by @fontawesome - https://fontawesome.com License - https://fontawesome.com/license/free (Icons: CC BY 4.0, Fonts: SIL OFL 1.1, Code: MIT License) Copyright 2025 Fonticons, Inc.--><path d="M439.8 200.5c-7.7-30.9-22.3-54.2-53.4-54.2h-40.1v47.4c0 36.8-31.2 67.8-66.8 67.8H172.7c-29.2 0-53.4 25-53.4 54.3v101.8c0 29 25.2 46 53.4 54.3 33.8 9.9 66.3 11.7 106.8 0 26.9-7.8 53.4-23.5 53.4-54.3v-40.7H226.2v-13.6h160.2c31.1 0 42.6-21.7 53.4-54.2 11.2-33.5 10.7-65.7 0-108.6M286.2 444.7a20.4 20.4 0 1 1 0-40.7 20.4 20.4 0 1 1 0 40.7M167.8 248.1h106.8c29.7 0 53.4-24.5 53.4-54.3V91.9c0-29-24.4-50.7-53.4-55.6-35.8-5.9-74.7-5.6-106.8.1-45.2 8-53.4 24.7-53.4 55.6v40.7h106.9v13.6h-147c-31.1 0-58.3 18.7-66.8 54.2-9.8 40.7-10.2 66.1 0 108.6 7.6 31.6 25.7 54.2 56.8 54.2H101v-48.8c0-35.3 30.5-66.4 66.8-66.4m-6.6-183.4a20.4 20.4 0 1 1 0 40.8 20.4 20.4 0 1 1 0-40.8"/></svg>
|
|
</a>
|
|
|
|
</div>
|
|
|
|
</div>
|
|
</div>
|
|
</footer>
|
|
|
|
</div>
|
|
<div class="md-dialog" data-md-component="dialog">
|
|
<div class="md-dialog__inner md-typeset"></div>
|
|
</div>
|
|
|
|
|
|
|
|
|
|
|
|
<script id="__config" type="application/json">{"annotate": null, "base": "..", "features": ["content.code.copy", "content.tabs.link", "navigation.instant", "navigation.tabs", "navigation.sections", "navigation.top", "search.highlight", "search.suggest", "toc.follow"], "search": "../assets/javascripts/workers/search.2c215733.min.js", "tags": null, "translations": {"clipboard.copied": "Copied to clipboard", "clipboard.copy": "Copy to clipboard", "search.result.more.one": "1 more on this page", "search.result.more.other": "# more on this page", "search.result.none": "No matching documents", "search.result.one": "1 matching document", "search.result.other": "# matching documents", "search.result.placeholder": "Type to start searching", "search.result.term.missing": "Missing", "select.version": "Select version"}, "version": null}</script>
|
|
|
|
|
|
<script src="../assets/javascripts/bundle.79ae519e.min.js"></script>
|
|
|
|
|
|
</body>
|
|
</html> |