apps/FastDeploy
1
0
Fork 0
mirror of https://github.com/PaddlePaddle/FastDeploy.git synced 2026-04-23 00:17:25 +08:00
Code Issues Actions 19 Packages Projects Releases Wiki Activity
Files
007c3ad0faf067fa1cf9b6d1d44e30033fbcb57a
FastDeploy/features/weight_update/index.html
T

3139 lines
59 KiB
HTML
Raw Blame History

<!doctype html>
<html lang="en" class="no-js">
<head>
<meta charset="utf-8">
<meta name="viewport" content="width=device-width,initial-scale=1">
<link rel="alternate" href="./" hreflang="en">
<link rel="alternate" href="../../zh/features/weight_update/" hreflang="zh">
<link rel="icon" href="../../assets/images/favicon.ico">
<meta name="generator" content="mkdocs-1.6.1, mkdocs-material-9.7.6">
<title>Weight update - FastDeploy: Large Language Model Deployment</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="#weight-clear-and-update" class="md-skip">
Skip to content
</a>
</div>
<div data-md-component="announce">
</div>
<header class="md-header md-header--shadow" data-md-component="header">
<nav class="md-header__inner md-grid" aria-label="Header">
<a href="../.." title="FastDeploy: Large Language Model Deployment" class="md-header__button md-logo" aria-label="FastDeploy: Large Language Model Deployment" data-md-component="logo">
<img src="../../assets/images/logo.jpg" alt="logo">
</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">
FastDeploy: Large Language Model Deployment
</span>
</div>
<div class="md-header__topic" data-md-component="header-topic">
<span class="md-ellipsis">
Weight update
</span>
</div>
</div>
</div>
<form class="md-header__option" data-md-component="palette">
<input class="md-option" data-md-color-media="(prefers-color-scheme: light)" 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="(prefers-color-scheme: dark)" data-md-color-scheme="slate" data-md-color-primary="black" data-md-color-accent="indigo" aria-label="Switch to system preference" type="radio" name="__palette" id="__palette_1">
<label class="md-header__button md-icon" title="Switch to system preference" 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>
<div class="md-header__option">
<div class="md-select">
<button class="md-header__button md-icon" aria-label="Select language">
<svg xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path d="m12.87 15.07-2.54-2.51.03-.03A17.5 17.5 0 0 0 14.07 6H17V4h-7V2H8v2H1v2h11.17C11.5 7.92 10.44 9.75 9 11.35 8.07 10.32 7.3 9.19 6.69 8h-2c.73 1.63 1.73 3.17 2.98 4.56l-5.09 5.02L4 19l5-5 3.11 3.11zM18.5 10h-2L12 22h2l1.12-3h4.75L21 22h2zm-2.62 7 1.62-4.33L19.12 17z"/></svg>
</button>
<div class="md-select__inner">
<ul class="md-select__list">
<li class="md-select__item">
<a href="./" hreflang="en" class="md-select__link">
English
</a>
</li>
<li class="md-select__item">
<a href="../../zh/features/weight_update/" hreflang="zh" class="md-select__link">
简体中文
</a>
</li>
</ul>
</div>
</div>
</div>
<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>
</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/PaddlePaddle/FastDeploy" 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 0 448 512"><!--! 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="M439.6 236.1 244 40.5c-5.4-5.5-12.8-8.5-20.4-8.5s-15 3-20.4 8.4L162.5 81l51.5 51.5c27.1-9.1 52.7 16.8 43.4 43.7l49.7 49.7c34.2-11.8 61.2 31 35.5 56.7-26.5 26.5-70.2-2.9-56-37.3L240.3 199v121.9c25.3 12.5 22.3 41.8 9.1 55-6.4 6.4-15.2 10.1-24.3 10.1s-17.8-3.6-24.3-10.1c-17.6-17.6-11.1-46.9 11.2-56v-123c-20.8-8.5-24.6-30.7-18.6-45L142.6 101 8.5 235.1C3 240.6 0 247.9 0 255.5s3 15 8.5 20.4l195.6 195.7c5.4 5.4 12.7 8.4 20.4 8.4s15-3 20.4-8.4l194.7-194.7c5.4-5.4 8.4-12.8 8.4-20.4s-3-15-8.4-20.4"/></svg>
</div>
<div class="md-source__repository">
FastDeploy
</div>
</a>
</div>
</nav>
</header>
<div class="md-container" data-md-component="container">
<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" aria-label="Navigation" data-md-level="0">
<label class="md-nav__title" for="__drawer">
<a href="../.." title="FastDeploy: Large Language Model Deployment" class="md-nav__button md-logo" aria-label="FastDeploy: Large Language Model Deployment" data-md-component="logo">
<img src="../../assets/images/logo.jpg" alt="logo">
</a>
FastDeploy: Large Language Model Deployment
</label>
<div class="md-nav__source">
<a href="https://github.com/PaddlePaddle/FastDeploy" 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 0 448 512"><!--! 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="M439.6 236.1 244 40.5c-5.4-5.5-12.8-8.5-20.4-8.5s-15 3-20.4 8.4L162.5 81l51.5 51.5c27.1-9.1 52.7 16.8 43.4 43.7l49.7 49.7c34.2-11.8 61.2 31 35.5 56.7-26.5 26.5-70.2-2.9-56-37.3L240.3 199v121.9c25.3 12.5 22.3 41.8 9.1 55-6.4 6.4-15.2 10.1-24.3 10.1s-17.8-3.6-24.3-10.1c-17.6-17.6-11.1-46.9 11.2-56v-123c-20.8-8.5-24.6-30.7-18.6-45L142.6 101 8.5 235.1C3 240.6 0 247.9 0 255.5s3 15 8.5 20.4l195.6 195.7c5.4 5.4 12.7 8.4 20.4 8.4s15-3 20.4-8.4l194.7-194.7c5.4-5.4 8.4-12.8 8.4-20.4s-3-15-8.4-20.4"/></svg>
</div>
<div class="md-source__repository">
FastDeploy
</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">
FastDeploy
</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">
Quick Start
</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>
Quick Start
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item md-nav__item--nested">
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_2_1" >
<label class="md-nav__link" for="__nav_2_1" id="__nav_2_1_label" tabindex="0">
<span class="md-ellipsis">
Installation
</span>
<span class="md-nav__icon md-icon"></span>
</label>
<nav class="md-nav" data-md-level="2" aria-labelledby="__nav_2_1_label" aria-expanded="false">
<label class="md-nav__title" for="__nav_2_1">
<span class="md-nav__icon md-icon"></span>
Installation
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../../get_started/installation/nvidia_gpu/" class="md-nav__link">
<span class="md-ellipsis">
Nvidia GPU
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/installation/kunlunxin_xpu/" class="md-nav__link">
<span class="md-ellipsis">
KunlunXin XPU
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/installation/intel_gaudi/" class="md-nav__link">
<span class="md-ellipsis">
Intel Gaudi
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/installation/hygon_dcu/" class="md-nav__link">
<span class="md-ellipsis">
HYGON DCU
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/installation/Enflame_gcu/" class="md-nav__link">
<span class="md-ellipsis">
Enflame S60
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/installation/iluvatar_gpu/" class="md-nav__link">
<span class="md-ellipsis">
Iluvatar CoreX
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/installation/metax_gpu/" class="md-nav__link">
<span class="md-ellipsis">
Metax C550
</span>
</a>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a href="../../get_started/quick_start/" class="md-nav__link">
<span class="md-ellipsis">
Quick Deployment For ERNIE-4.5-0.3B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/quick_start_vl/" class="md-nav__link">
<span class="md-ellipsis">
Quick Deployment for ERNIE-4.5-VL-28B-A3B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/ernie-4.5/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-300B-A47B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/ernie-4.5-vl/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-VL-424B-A47B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/quick_start_qwen/" class="md-nav__link">
<span class="md-ellipsis">
Quick Deployment For QWEN
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../get_started/quick_start_qwen25_vl/" class="md-nav__link">
<span class="md-ellipsis">
Quick Deployment For QWEN2.5-VL
</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">
Online Serving
</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>
Online Serving
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../../online_serving/" class="md-nav__link">
<span class="md-ellipsis">
OpenAI-Compatible API Server
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../online_serving/metrics/" class="md-nav__link">
<span class="md-ellipsis">
Monitor Metrics
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../online_serving/scheduler/" class="md-nav__link">
<span class="md-ellipsis">
Scheduler
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../online_serving/graceful_shutdown_service/" class="md-nav__link">
<span class="md-ellipsis">
Graceful Shutdown
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../online_serving/router/" class="md-nav__link">
<span class="md-ellipsis">
Load-Balancing Scheduling Router
</span>
</a>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a href="../../offline_inference/" class="md-nav__link">
<span class="md-ellipsis">
Offline Inference
</span>
</a>
</li>
<li class="md-nav__item md-nav__item--nested">
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_5" >
<label class="md-nav__link" for="__nav_5" id="__nav_5_label" tabindex="0">
<span class="md-ellipsis">
Best Practices
</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="false">
<label class="md-nav__title" for="__nav_5">
<span class="md-nav__icon md-icon"></span>
Best Practices
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../../best_practices/ERNIE-4.5-0.3B-Paddle/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-0.3B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../best_practices/ERNIE-4.5-21B-A3B-Paddle/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-21B-A3B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../best_practices/ERNIE-4.5-300B-A47B-Paddle/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-300B-A47B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../best_practices/ERNIE-4.5-21B-A3B-Thinking/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-21B-A3B-Thinking
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../best_practices/ERNIE-4.5-VL-28B-A3B-Paddle/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-VL-28B-A3B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../best_practices/ERNIE-4.5-VL-424B-A47B-Paddle/" class="md-nav__link">
<span class="md-ellipsis">
ERNIE-4.5-VL-424B-A47B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../best_practices/PaddleOCR-VL-0.9B/" class="md-nav__link">
<span class="md-ellipsis">
PaddleOCR-VL-0.9B
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../best_practices/FAQ/" class="md-nav__link">
<span class="md-ellipsis">
FAQ
</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">
Quantization
</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>
Quantization
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../../quantization/" class="md-nav__link">
<span class="md-ellipsis">
Overview
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../quantization/online_quantization/" class="md-nav__link">
<span class="md-ellipsis">
Online Quantization
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../quantization/wint2/" class="md-nav__link">
<span class="md-ellipsis">
WINT2 Quantization
</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">
Features
</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>
Features
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../prefix_caching/" class="md-nav__link">
<span class="md-ellipsis">
Prefix Caching
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../disaggregated/" class="md-nav__link">
<span class="md-ellipsis">
Disaggregation
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../chunked_prefill/" class="md-nav__link">
<span class="md-ellipsis">
Chunked Prefill
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../load_balance/" class="md-nav__link">
<span class="md-ellipsis">
Load Balance
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../speculative_decoding/" class="md-nav__link">
<span class="md-ellipsis">
Speculative Decoding
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../structured_outputs/" class="md-nav__link">
<span class="md-ellipsis">
Structured Outputs
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../reasoning_output/" class="md-nav__link">
<span class="md-ellipsis">
Reasoning Output
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../thinking_budget/" class="md-nav__link">
<span class="md-ellipsis">
Thinking Budget
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../early_stop/" class="md-nav__link">
<span class="md-ellipsis">
Early Stop
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../plugins/" class="md-nav__link">
<span class="md-ellipsis">
Plugins
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../paddleformers_backend/" class="md-nav__link">
<span class="md-ellipsis">
PaddleFormers Backend
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../sampling/" class="md-nav__link">
<span class="md-ellipsis">
Sampling
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../multi-node_deployment/" class="md-nav__link">
<span class="md-ellipsis">
MultiNode Deployment
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../graph_optimization/" class="md-nav__link">
<span class="md-ellipsis">
Graph Optimization
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../data_parallel_service/" class="md-nav__link">
<span class="md-ellipsis">
Data Parallelism
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../plas_attention/" class="md-nav__link">
<span class="md-ellipsis">
PLAS
</span>
</a>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a href="../../supported_models/" class="md-nav__link">
<span class="md-ellipsis">
Supported Models
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../benchmark/" class="md-nav__link">
<span class="md-ellipsis">
Benchmark
</span>
</a>
</li>
<li class="md-nav__item md-nav__item--nested">
<input class="md-nav__toggle md-toggle " type="checkbox" id="__nav_10" >
<label class="md-nav__link" for="__nav_10" id="__nav_10_label" tabindex="0">
<span class="md-ellipsis">
Usage
</span>
<span class="md-nav__icon md-icon"></span>
</label>
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_10_label" aria-expanded="false">
<label class="md-nav__title" for="__nav_10">
<span class="md-nav__icon md-icon"></span>
Usage
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../../usage/log/" class="md-nav__link">
<span class="md-ellipsis">
Log Description
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../usage/code_overview/" class="md-nav__link">
<span class="md-ellipsis">
Code Overview
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../usage/environment_variables/" class="md-nav__link">
<span class="md-ellipsis">
Environment Variables
</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_11" >
<label class="md-nav__link" for="__nav_11" id="__nav_11_label" tabindex="0">
<span class="md-ellipsis">
CLI
</span>
<span class="md-nav__icon md-icon"></span>
</label>
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_11_label" aria-expanded="false">
<label class="md-nav__title" for="__nav_11">
<span class="md-nav__icon md-icon"></span>
CLI
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../../cli/" class="md-nav__link">
<span class="md-ellipsis">
Overview
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../cli/chat/" class="md-nav__link">
<span class="md-ellipsis">
Chat
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../cli/complete/" class="md-nav__link">
<span class="md-ellipsis">
Complete
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../cli/serve/" class="md-nav__link">
<span class="md-ellipsis">
Server
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../cli/collect-env/" class="md-nav__link">
<span class="md-ellipsis">
Collect Env
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../cli/bench/" class="md-nav__link">
<span class="md-ellipsis">
Bench
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../cli/run-batch/" class="md-nav__link">
<span class="md-ellipsis">
Run Batch
</span>
</a>
</li>
<li class="md-nav__item">
<a href="../../cli/tokenizer/" class="md-nav__link">
<span class="md-ellipsis">
Tokenizer
</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_12" >
<label class="md-nav__link" for="__nav_12" id="__nav_12_label" tabindex="0">
<span class="md-ellipsis">
Observability
</span>
<span class="md-nav__icon md-icon"></span>
</label>
<nav class="md-nav" data-md-level="1" aria-labelledby="__nav_12_label" aria-expanded="false">
<label class="md-nav__title" for="__nav_12">
<span class="md-nav__icon md-icon"></span>
Observability
</label>
<ul class="md-nav__list" data-md-scrollfix>
<li class="md-nav__item">
<a href="../../observability/trace/" class="md-nav__link">
<span class="md-ellipsis">
Trace
</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="#prerequisites" class="md-nav__link">
<span class="md-ellipsis">
Prerequisites
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#api-overview" class="md-nav__link">
<span class="md-ellipsis">
API Overview
</span>
</a>
<nav class="md-nav" aria-label="API Overview">
<ul class="md-nav__list">
<li class="md-nav__item">
<a href="#compatibility-apis" class="md-nav__link">
<span class="md-ellipsis">
Compatibility APIs
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#v1-control-apis" class="md-nav__link">
<span class="md-ellipsis">
V1 control APIs
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#compatibility-notes" class="md-nav__link">
<span class="md-ellipsis">
Compatibility Notes
</span>
</a>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a href="#interface-semantics" class="md-nav__link">
<span class="md-ellipsis">
Interface Semantics
</span>
</a>
<nav class="md-nav" aria-label="Interface Semantics">
<ul class="md-nav__list">
<li class="md-nav__item">
<a href="#v1pause" class="md-nav__link">
<span class="md-ellipsis">
/v1/pause
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#v1sleep" class="md-nav__link">
<span class="md-ellipsis">
/v1/sleep
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#v1wakeup" class="md-nav__link">
<span class="md-ellipsis">
/v1/wakeup
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#v1update_weights" class="md-nav__link">
<span class="md-ellipsis">
/v1/update_weights
</span>
</a>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a href="#example-requests" class="md-nav__link">
<span class="md-ellipsis">
Example Requests
</span>
</a>
<nav class="md-nav" aria-label="Example Requests">
<ul class="md-nav__list">
<li class="md-nav__item">
<a href="#basic-apis" class="md-nav__link">
<span class="md-ellipsis">
Basic APIs
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#sleep-wakeup-apis" class="md-nav__link">
<span class="md-ellipsis">
Sleep / Wakeup APIs
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#update-weights-api" class="md-nav__link">
<span class="md-ellipsis">
Update Weights API
</span>
</a>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a href="#rlhf-usage" class="md-nav__link">
<span class="md-ellipsis">
RLHF Usage
</span>
</a>
<nav class="md-nav" aria-label="RLHF Usage">
<ul class="md-nav__list">
<li class="md-nav__item">
<a href="#recommended-rollout-service-setup" class="md-nav__link">
<span class="md-ellipsis">
Recommended Rollout Service Setup
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#training-side-integration-support" class="md-nav__link">
<span class="md-ellipsis">
Training-Side Integration Support
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#common-rlhf-workflows" class="md-nav__link">
<span class="md-ellipsis">
Common RLHF workflows
</span>
</a>
</li>
</ul>
</nav>
</li>
<li class="md-nav__item">
<a href="#other-related-configuration" class="md-nav__link">
<span class="md-ellipsis">
Other Related Configuration
</span>
</a>
<nav class="md-nav" aria-label="Other Related Configuration">
<ul class="md-nav__list">
<li class="md-nav__item">
<a href="#communication-group-clear-and-rebuild" class="md-nav__link">
<span class="md-ellipsis">
Communication Group Clear and Rebuild
</span>
</a>
</li>
<li class="md-nav__item">
<a href="#cpu-cache-clear-and-rebuild" class="md-nav__link">
<span class="md-ellipsis">
CPU Cache Clear and Rebuild
</span>
</a>
</li>
</ul>
</nav>
</li>
</ul>
</nav>
</div>
</div>
</div>
<div class="md-content" data-md-component="content">
<article class="md-content__inner md-typeset">
<p><a href="../zh/features/weight_update.md">简体中文</a></p>
<h1 id="weight-clear-and-update">Weight Clear and Update</h1>
<p>FastDeploy supports dynamic weight clear and update for RL and RLHF rollout services. This capability is primarily intended to address the following two requirements:</p>
<ul>
<li>release GPU memory when the rollout engine is idle;</li>
<li>refresh inference weights after the trainer produces a new checkpoint, without restarting the whole service.</li>
</ul>
<p>This page describes the weight-control interfaces currently supported by FastDeploy, the semantics of each interface, and their typical usage in RLHF training.</p>
<h2 id="prerequisites">Prerequisites</h2>
<p>In RLHF scenarios, FastDeploy mainly provides this capability through the online serving mode. Dynamic weight loading must be enabled when starting the service:</p>
<pre><code class="language-bash">python -m fastdeploy.entrypoints.openai.api_server \
--model /path/to/model \
--dynamic-load-weight \
--load_strategy ipc_snapshot
</code></pre>
<p><code>--dynamic-load-weight</code> enables dynamic weight control, and <code>--load_strategy</code> specifies the concrete weight update mechanism. The currently supported update modes are listed below:</p>
<table>
<thead>
<tr>
<th>Mode</th>
<th><code>load_strategy</code></th>
<th>Typical use</th>
<th>Notes</th>
</tr>
</thead>
<tbody>
<tr>
<td>CUDA IPC</td>
<td><code>ipc</code></td>
<td>Training and inference processes on the same node share live tensors</td>
<td>Update source comes from IPC metadata produced by the training side.</td>
</tr>
<tr>
<td>IPC snapshot</td>
<td><code>ipc_snapshot</code></td>
<td>Rollout reloads a snapshot file produced by training</td>
<td>Used by current RL rollout examples.</td>
</tr>
<tr>
<td>RDMA / rsync</td>
<td><code>rsync</code></td>
<td>Trainer publishes a new version and rollout fetches it remotely</td>
<td><code>POST /v1/update_weights</code> is the explicit API for this mode.</td>
</tr>
</tbody>
</table>
<h2 id="api-overview">API Overview</h2>
<h3 id="compatibility-apis">Compatibility APIs</h3>
<p>In FastDeploy &lt;= 2.5, the following simplified APIs are provided for compatibility with the legacy RL control flow.</p>
<table>
<thead>
<tr>
<th>API</th>
<th>Method</th>
<th>Meaning</th>
<th>Availability</th>
</tr>
</thead>
<tbody>
<tr>
<td><code>/clear_load_weight</code></td>
<td><code>GET</code></td>
<td>Clear or offload currently loaded weights</td>
<td>Requires <code>dynamic_load_weight=True</code></td>
</tr>
<tr>
<td><code>/update_model_weight</code></td>
<td><code>GET</code></td>
<td>Reload weights after a clear/offload operation</td>
<td>Requires <code>dynamic_load_weight=True</code></td>
</tr>
</tbody>
</table>
<h3 id="v1-control-apis">V1 control APIs</h3>
<p>In FastDeploy &gt;= 2.6, the underlying control-signal communication path is optimized and V1 control APIs are introduced. Compared with the legacy APIs, the V1 APIs provide a more stable execution path, clearer semantics, and more flexible control:</p>
<table>
<thead>
<tr>
<th>API</th>
<th>Method</th>
<th>Request params</th>
<th>Semantics</th>
</tr>
</thead>
<tbody>
<tr>
<td><code>/v1/pause</code></td>
<td><code>POST</code></td>
<td>none</td>
<td>Pause request generation, abort running and inflight requests, reset scheduler state, and pause cache transfer if enabled.</td>
</tr>
<tr>
<td><code>/v1/resume</code></td>
<td><code>POST</code></td>
<td>none</td>
<td>Resume request generation and cache transfer.</td>
</tr>
<tr>
<td><code>/v1/is_paused</code></td>
<td><code>GET</code></td>
<td>none</td>
<td>Return <code>{"is_paused": bool}</code>.</td>
</tr>
<tr>
<td><code>/v1/sleep</code></td>
<td><code>POST</code></td>
<td><code>?tags=weight,kv_cache</code></td>
<td>Offload selected GPU memory objects. Supported tags are <code>weight</code> and <code>kv_cache</code>. If omitted, both are used.</td>
</tr>
<tr>
<td><code>/v1/wakeup</code></td>
<td><code>POST</code></td>
<td><code>?tags=weight,kv_cache</code></td>
<td>Reload previously offloaded weights and/or KV cache. On success, the engine resumes automatically.</td>
</tr>
<tr>
<td><code>/v1/update_weights</code></td>
<td><code>POST</code></td>
<td>JSON <code>{"version":"...", "verify_checksum": false}</code></td>
<td>Refresh weights in place through the worker control path. This API is intended for remote versioned updates, especially <code>load_strategy=rsync</code>.</td>
</tr>
</tbody>
</table>
<h3 id="compatibility-notes">Compatibility Notes</h3>
<p>The optimized communication path also applies to the legacy APIs. By setting <code>FD_ENABLE_V1_UPDATE_WEIGHTS=1</code>, the legacy APIs can be switched to the new control path while keeping the original API form.</p>
<ul>
<li><code>FD_ENABLE_V1_UPDATE_WEIGHTS=0</code>: use the legacy shared-memory-based control path.</li>
<li><code>FD_ENABLE_V1_UPDATE_WEIGHTS=1</code>: <code>/clear_load_weight</code> is effectively handled through <code>/v1/sleep</code>, and <code>/update_model_weight</code> is effectively handled through <code>/v1/wakeup</code>. The corresponding pause/resume actions are handled internally by <code>sleep</code> and <code>wakeup</code>.</li>
</ul>
<p><strong>Note</strong>: regardless of whether V1 is enabled, the legacy APIs are not the recommended standard interface for RLHF scenarios and may be gradually deprecated in future releases. The <code>/v1/*</code> control APIs are recommended.</p>
<h2 id="interface-semantics">Interface Semantics</h2>
<h3 id="v1pause"><code>/v1/pause</code></h3>
<p><code>/v1/pause</code> is the safe boundary before changing model state.</p>
<p>It does the following:</p>
<ul>
<li>stops new request generation;</li>
<li>aborts running and inflight requests;</li>
<li>resets scheduler state;</li>
<li>pauses cache transfer when multi-level cache or KV cache storage is enabled.</li>
</ul>
<p>When a clear boundary is required between one rollout round and the next training stage, this API should be called first.</p>
<h3 id="v1sleep"><code>/v1/sleep</code></h3>
<p><code>/v1/sleep</code> offloads selected runtime state from GPU memory.</p>
<p>Supported tags:</p>
<ul>
<li><code>weight</code>: clear model weights from device memory; if enabled, communication groups and DeepEP buffers may also be released.</li>
<li><code>kv_cache</code>: clear KV cache; MTP cache is also cleared when speculative decoding uses MTP.</li>
</ul>
<p>If the <code>tags</code> parameter is omitted, FastDeploy defaults to:</p>
<pre><code class="language-bash">/v1/sleep?tags=weight,kv_cache
</code></pre>
<p>In the current implementation, <code>sleep</code> automatically performs a <code>pause</code> first. New integrations should not rely on this implicit behavior.</p>
<h3 id="v1wakeup"><code>/v1/wakeup</code></h3>
<p><code>/v1/wakeup</code> restores the state offloaded by <code>/v1/sleep</code>.</p>
<p>Depending on tags and configuration, FastDeploy may:</p>
<ul>
<li>restart communication groups;</li>
<li>recreate DeepEP buffers;</li>
<li>reload model weights from the configured source;</li>
<li>rebuild KV cache;</li>
<li>recapture CUDA Graph.</li>
</ul>
<p>After <code>wakeup</code> succeeds, FastDeploy automatically calls <code>resume</code>.</p>
<h3 id="v1update_weights"><code>/v1/update_weights</code></h3>
<p><code>/v1/update_weights</code> refreshes model parameters directly, without unloading the GPU memory occupied by model weights.</p>
<p>Current request fields:</p>
<ul>
<li><code>version</code>: optional string. Used to choose a target checkpoint version.</li>
<li><code>verify_checksum</code>: optional boolean. Defaults to <code>false</code>. Set to <code>true</code> to verify data integrity during weight synchronization.</li>
</ul>
<p>Important semantics:</p>
<ul>
<li>the engine must already be paused, otherwise the request fails;</li>
<li>the update is executed on workers only;</li>
<li>this API is meant for explicit weight refresh, especially the <code>rsync</code> path;</li>
<li>it does not implicitly call <code>resume</code>.</li>
</ul>
<p>Recommended sequence:</p>
<ol>
<li><code>POST /v1/pause</code></li>
<li><code>POST /v1/update_weights</code></li>
<li><code>POST /v1/resume</code></li>
</ol>
<p>If GPU memory also needs to be reclaimed between rollout rounds, the <code>sleep</code> / <code>wakeup</code> workflow is more appropriate.</p>
<h2 id="example-requests">Example Requests</h2>
<h3 id="basic-apis">Basic APIs</h3>
<p>Pause the engine:</p>
<pre><code class="language-bash">curl -X POST http://127.0.0.1:8000/v1/pause
</code></pre>
<p>Resume the engine:</p>
<pre><code class="language-bash">curl -X POST http://127.0.0.1:8000/v1/resume
</code></pre>
<h3 id="sleep-wakeup-apis">Sleep / Wakeup APIs</h3>
<p><strong>Offload weights and KV cache</strong></p>
<pre><code class="language-bash"># Offload both weights and KV cache
curl -X POST &quot;http://127.0.0.1:8000/v1/sleep?tags=weight,kv_cache&quot;
# Offload only weights
curl -X POST &quot;http://127.0.0.1:8000/v1/sleep?tags=weight&quot;
# Omit parameter, defaults to both
curl -X POST &quot;http://127.0.0.1:8000/v1/sleep&quot;
</code></pre>
<p><strong>Restore weights and KV cache</strong></p>
<pre><code class="language-bash"># Restore both weights and KV cache
curl -X POST &quot;http://127.0.0.1:8000/v1/wakeup?tags=weight,kv_cache&quot;
# Restore only weights
curl -X POST &quot;http://127.0.0.1:8000/v1/wakeup?tags=weight&quot;
# Omit parameter, defaults to both
curl -X POST &quot;http://127.0.0.1:8000/v1/wakeup&quot;
</code></pre>
<p><strong>Note</strong>: When <code>use_cudagraph=True</code>, KV cache must be restored before weights. This means <code>/v1/wakeup</code> with the <code>kv_cache</code> tag must be called before calling <code>/v1/wakeup</code> with the <code>weight</code> tag. If weights are restored without KV cache, an error will be raised. It is recommended to keep the <code>tags</code> parameter consistent between <code>/v1/sleep</code> and <code>/v1/wakeup</code>.</p>
<h3 id="update-weights-api">Update Weights API</h3>
<p>Refresh to a new remotely published version:</p>
<pre><code class="language-bash">curl -X POST http://127.0.0.1:8000/v1/update_weights \
-H &quot;Content-Type: application/json&quot; \
-d '{
&quot;version&quot;: &quot;global_step_1200&quot;,
&quot;verify_checksum&quot;: false
}'
</code></pre>
<h2 id="rlhf-usage">RLHF Usage</h2>
<h3 id="recommended-rollout-service-setup">Recommended Rollout Service Setup</h3>
<p>In RLHF scenarios, FastDeploy rollout services are typically configured as follows:</p>
<ul>
<li><code>dynamic_load_weight=True</code></li>
<li><code>load_strategy=ipc_snapshot</code> for local snapshot-based refresh;</li>
<li>or <code>load_strategy=rsync</code> for versioned remote refresh.</li>
</ul>
<p>The rollout utilities in the repository already follow this pattern. A typical example is:</p>
<pre><code class="language-python">from fastdeploy.rl.rollout_config import RolloutModelConfig
from fastdeploy.rl.rollout_model import RolloutModel
rollout_config = RolloutModelConfig(
model_name_or_path=model_path,
tensor_parallel_size=ranks,
dynamic_load_weight=True,
load_strategy=&quot;ipc_snapshot&quot;,
)
rollout_model = RolloutModel(rollout_config)
</code></pre>
<h3 id="training-side-integration-support">Training-Side Integration Support</h3>
<p>In addition to serving endpoints, FastDeploy provides the following training-side integration capabilities for RLHF:</p>
<ul>
<li><code>RolloutModel.state_dict()</code>: exposes the rollout-side inference parameters.</li>
<li><code>RolloutModel.get_name_mappings_to_training()</code>: exposes the mapping from inference parameter names to training parameter names.</li>
</ul>
<p>These interfaces can be used to align training checkpoints with rollout-side parameter layouts, especially when inference-side and training-side parameter names are not fully identical.</p>
<h3 id="common-rlhf-workflows">Common RLHF workflows</h3>
<p>The following examples assume the service endpoint is <code>http://127.0.0.1:8000</code>.</p>
<p><strong>Workflow 1: clear and restore</strong></p>
<p>This workflow is suitable when the rollout service stays resident, but GPU memory should be released before training and restored afterward. The recommended sequence is <code>(pause) -&gt; sleep -&gt; wakeup -&gt; (resume)</code>, where the steps in parentheses are optional.</p>
<pre><code class="language-bash"># Optional: explicitly pause the engine to establish a clear transition boundary
curl -X POST http://127.0.0.1:8000/v1/pause
# Offload both weights and KV cache
curl -X POST &quot;http://127.0.0.1:8000/v1/sleep?tags=weight,kv_cache&quot;
# Restore both weights and KV cache after training completes
curl -X POST &quot;http://127.0.0.1:8000/v1/wakeup?tags=weight,kv_cache&quot;
# Optional: explicitly resume if required by the integration
curl -X POST http://127.0.0.1:8000/v1/resume
</code></pre>
<p><strong>Workflow 2: in-place refresh to a new checkpoint</strong></p>
<p>This workflow is suitable when the service remains resident and only needs to switch to a new checkpoint version. The recommended sequence is <code>pause -&gt; update_weights -&gt; resume</code>.</p>
<pre><code class="language-bash"># Pause the engine first
curl -X POST http://127.0.0.1:8000/v1/pause
# Refresh to a new checkpoint version in place
curl -X POST http://127.0.0.1:8000/v1/update_weights \
-H &quot;Content-Type: application/json&quot; \
-d '{
&quot;version&quot;: &quot;global_step_1200&quot;,
&quot;verify_checksum&quot;: false
}'
# Resume the service after the update completes
curl -X POST http://127.0.0.1:8000/v1/resume
</code></pre>
<p><strong>Workflow 3: legacy compatibility APIs</strong></p>
<p>Legacy RL clients can continue to use the compatibility flow <code>clear_load_weight -&gt; update_model_weight</code>.</p>
<pre><code class="language-bash"># Clear or offload the current weights
curl -X GET http://127.0.0.1:8000/clear_load_weight
# Reload weights after the trainer updates the checkpoint
curl -X GET http://127.0.0.1:8000/update_model_weight
</code></pre>
<p>For new integrations, the <code>/v1/*</code> APIs are recommended because their control path is more explicit and easier to trace.</p>
<h2 id="other-related-configuration">Other Related Configuration</h2>
<h3 id="communication-group-clear-and-rebuild">Communication Group Clear and Rebuild</h3>
<p>FastDeploy provides <code>--shutdown-comm-group-if-worker-idle</code> and <code>--no-shutdown-comm-group-if-worker-idle</code> to explicitly control whether communication groups should also be torn down when weights are offloaded.</p>
<p>Keeping communication groups alive generally improves the stability of weight clearing and reloading. The tradeoff is that more GPU memory remains allocated after weight offload, and the execution time of <code>sleep</code> / <code>wakeup</code> may also increase.</p>
<p>By default:</p>
<ul>
<li>in EP scenarios, communication groups are kept;</li>
<li>in non-EP scenarios, communication groups are torn down.</li>
</ul>
<h3 id="cpu-cache-clear-and-rebuild">CPU Cache Clear and Rebuild</h3>
<p>After <code>--swap-space</code> is enabled, the following environment variable can be used to control whether CPU-side cache should also be cleared when <code>/v1/sleep</code> is executed, in order to reduce memory pressure during training.</p>
<p>By default, FastDeploy does not actively clear CPU cache. To clear it together with <code>sleep</code>, set:</p>
<pre><code class="language-bash">export FD_ENABLE_SWAP_SPACE_CLEARING=1
</code></pre>
</article>
</div>
<script>var target=document.getElementById(location.hash.slice(1));target&&target.name&&(target.checked=target.name.startsWith("__tabbed_"))</script>
</div>
</main>
<footer class="md-footer">
<div class="md-footer-meta md-typeset">
<div class="md-footer-meta__inner md-grid">
<div class="md-copyright">
<div class="md-copyright__highlight">
Copyright &copy; 2025 Maintained by FastDeploy
</div>
Made with
<a href="https://squidfunk.github.io/mkdocs-material/" target="_blank" rel="noopener">
Material for MkDocs
</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": [], "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>
Reference in New Issue View Git Blame Copy Permalink