CINXE.COM

<!doctype html> <html class=""> <head> <meta charset="utf-8" /> <meta name="viewport" content="width=device-width, initial-scale=1.0, user-scalable=no" /> <meta name="description" content="We’re on a journey to advance and democratize artificial intelligence through open source and open science." /> <meta property="fb:app_id" content="1321688464574422" /> <meta name="twitter:card" content="summary_large_image" /> <meta name="twitter:site" content="@huggingface" /> <meta name="twitter:image" content="https://huggingface.co/front/thumbnails/docs/trl.png" /> <meta property="og:title" content="Nash-MD Trainer" /> <meta property="og:type" content="website" /> <meta property="og:url" content="https://huggingface.co/docs/trl/v0.15.2/en/nash_md_trainer" /> <meta property="og:image" content="https://huggingface.co/front/thumbnails/docs/trl.png" /> <link rel="stylesheet" href="/front/build/kube-859986b/style.css" /> <link rel="preconnect" href="https://fonts.gstatic.com" /> <link href="https://fonts.googleapis.com/css2?family=Source+Sans+Pro:ital,wght@0,200;0,300;0,400;0,600;0,700;0,900;1,200;1,300;1,400;1,600;1,700;1,900&display=swap" rel="stylesheet" /> <link href="https://fonts.googleapis.com/css2?family=IBM+Plex+Mono:wght@400;600;700&display=swap" rel="stylesheet" /> <link rel="preload" href="https://cdnjs.cloudflare.com/ajax/libs/KaTeX/0.12.0/katex.min.css" as="style" onload="this.onload=null;this.rel='stylesheet'" /> <noscript> <link rel="stylesheet" href="https://cdnjs.cloudflare.com/ajax/libs/KaTeX/0.12.0/katex.min.css" /> </noscript> <script>const guestTheme = document.cookie.match(/theme=(\w+)/)?.[1]; document.documentElement.classList.toggle('dark', guestTheme === 'dark' || ( (!guestTheme || guestTheme === 'system') && window.matchMedia('(prefers-color-scheme: dark)').matches));</script> <link rel="canonical" href="https://huggingface.co/docs/trl/v0.15.2/en/nash_md_trainer"> <link rel="alternate" hreflang="en" href="https://huggingface.co/docs/trl/v0.15.2/en/nash_md_trainer"> <link rel="alternate" hreflang="x-default" href="https://huggingface.co/docs/trl/v0.15.2/nash_md_trainer"> <title>Nash-MD Trainer</title> <script defer data-domain="huggingface.co" event-loggedIn="false" src="/js/script.pageview-props.js" ></script> <script> window.plausible = window.plausible || function () { (window.plausible.q = window.plausible.q || []).push(arguments); }; </script> <script> window.hubConfig = {"features":{"signupDisabled":false},"sshGitUrl":"git@hf.co","moonHttpUrl":"https:\/\/huggingface.co","captchaApiKey":"bd5f2066-93dc-4bdd-a64b-a24646ca3859","captchaDisabledOnSignup":true,"datasetViewerPublicUrl":"https:\/\/datasets-server.huggingface.co","stripePublicKey":"pk_live_x2tdjFXBCvXo2FFmMybezpeM00J6gPCAAc","environment":"production","userAgent":"HuggingFace (production)","spacesIframeDomain":"hf.space","spacesApiUrl":"https:\/\/api.hf.space","docSearchKey":"ece5e02e57300e17d152c08056145326e90c4bff3dd07d7d1ae40cf1c8d39cb6","logoDev":{"apiUrl":"https:\/\/img.logo.dev\/","apiKey":"pk_UHS2HZOeRnaSOdDp7jbd5w"}}; </script> <script type="text/javascript" src="https://de5282c3ca0c.edge.sdk.awswaf.com/de5282c3ca0c/526cf06acb0d/challenge.js" defer></script> </head> <body class="flex flex-col min-h-dvh bg-white dark:bg-gray-950 text-black DocBuilderPage"> <div class="flex min-h-dvh flex-col"><div class="SVELTE_HYDRATER contents" data-target="SystemThemeMonitor" data-props="{"isLoggedIn":false}"></div> <div class="SVELTE_HYDRATER contents" data-target="MainHeader" data-props="{"classNames":"","isWide":true,"isZh":false,"isPro":false}"><header class="border-b border-gray-100 "><div class="w-full px-4 flex h-16 items-center"><div class="flex flex-1 items-center"><a class="mr-5 flex flex-none items-center lg:mr-6" href="/"><img alt="Hugging Face's logo" class="w-7 md:mr-2" src="/front/assets/huggingface_logo-noborder.svg"> <span class="hidden whitespace-nowrap text-lg font-bold md:block">Hugging Face</span></a> <div class="relative flex-1 lg:max-w-sm mr-2 sm:mr-4 md:mr-3 xl:mr-6"><input autocomplete="off" class="w-full dark:bg-gray-950 pl-8 form-input-alt h-9 pr-3 focus:shadow-xl " name="" placeholder="Search models, datasets, users..." spellcheck="false" type="text" value=""> <svg class="absolute left-2.5 text-gray-400 top-1/2 transform -translate-y-1/2" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M30 28.59L22.45 21A11 11 0 1 0 21 22.45L28.59 30zM5 14a9 9 0 1 1 9 9a9 9 0 0 1-9-9z" fill="currentColor"></path></svg> </div> <div class="flex flex-none items-center justify-center p-0.5 place-self-stretch lg:hidden"><button class="relative z-40 flex h-6 w-8 items-center justify-center" type="button"><svg width="1em" height="1em" viewBox="0 0 10 10" class="text-xl" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" preserveAspectRatio="xMidYMid meet" fill="currentColor"><path fill-rule="evenodd" clip-rule="evenodd" d="M1.65039 2.9999C1.65039 2.8066 1.80709 2.6499 2.00039 2.6499H8.00039C8.19369 2.6499 8.35039 2.8066 8.35039 2.9999C8.35039 3.1932 8.19369 3.3499 8.00039 3.3499H2.00039C1.80709 3.3499 1.65039 3.1932 1.65039 2.9999ZM1.65039 4.9999C1.65039 4.8066 1.80709 4.6499 2.00039 4.6499H8.00039C8.19369 4.6499 8.35039 4.8066 8.35039 4.9999C8.35039 5.1932 8.19369 5.3499 8.00039 5.3499H2.00039C1.80709 5.3499 1.65039 5.1932 1.65039 4.9999ZM2.00039 6.6499C1.80709 6.6499 1.65039 6.8066 1.65039 6.9999C1.65039 7.1932 1.80709 7.3499 2.00039 7.3499H8.00039C8.19369 7.3499 8.35039 7.1932 8.35039 6.9999C8.35039 6.8066 8.19369 6.6499 8.00039 6.6499H2.00039Z"></path></svg> </button> </div></div> <nav aria-label="Main" class="ml-auto hidden lg:block"><ul class="flex items-center space-x-1.5 2xl:space-x-2"><li class="hover:text-indigo-700"><a class="group flex items-center px-2 py-0.5 dark:text-gray-300 dark:hover:text-gray-100" href="/models"><svg class="mr-1.5 text-gray-400 group-hover:text-indigo-500" style="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24"><path class="uim-quaternary" d="M20.23 7.24L12 12L3.77 7.24a1.98 1.98 0 0 1 .7-.71L11 2.76c.62-.35 1.38-.35 2 0l6.53 3.77c.29.173.531.418.7.71z" opacity=".25" fill="currentColor"></path><path class="uim-tertiary" d="M12 12v9.5a2.09 2.09 0 0 1-.91-.21L4.5 17.48a2.003 2.003 0 0 1-1-1.73v-7.5a2.06 2.06 0 0 1 .27-1.01L12 12z" opacity=".5" fill="currentColor"></path><path class="uim-primary" d="M20.5 8.25v7.5a2.003 2.003 0 0 1-1 1.73l-6.62 3.82c-.275.13-.576.198-.88.2V12l8.23-4.76c.175.308.268.656.27 1.01z" fill="currentColor"></path></svg> Models</a> </li><li class="hover:text-red-700"><a class="group flex items-center px-2 py-0.5 dark:text-gray-300 dark:hover:text-gray-100" href="/datasets"><svg class="mr-1.5 text-gray-400 group-hover:text-red-500" style="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 25 25"><ellipse cx="12.5" cy="5" fill="currentColor" fill-opacity="0.25" rx="7.5" ry="2"></ellipse><path d="M12.5 15C16.6421 15 20 14.1046 20 13V20C20 21.1046 16.6421 22 12.5 22C8.35786 22 5 21.1046 5 20V13C5 14.1046 8.35786 15 12.5 15Z" fill="currentColor" opacity="0.5"></path><path d="M12.5 7C16.6421 7 20 6.10457 20 5V11.5C20 12.6046 16.6421 13.5 12.5 13.5C8.35786 13.5 5 12.6046 5 11.5V5C5 6.10457 8.35786 7 12.5 7Z" fill="currentColor" opacity="0.5"></path><path d="M5.23628 12C5.08204 12.1598 5 12.8273 5 13C5 14.1046 8.35786 15 12.5 15C16.6421 15 20 14.1046 20 13C20 12.8273 19.918 12.1598 19.7637 12C18.9311 12.8626 15.9947 13.5 12.5 13.5C9.0053 13.5 6.06886 12.8626 5.23628 12Z" fill="currentColor"></path></svg> Datasets</a> </li><li class="hover:text-blue-700"><a class="group flex items-center px-2 py-0.5 dark:text-gray-300 dark:hover:text-gray-100" href="/spaces"><svg class="mr-1.5 text-gray-400 group-hover:text-blue-500" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" viewBox="0 0 25 25"><path opacity=".5" d="M6.016 14.674v4.31h4.31v-4.31h-4.31ZM14.674 14.674v4.31h4.31v-4.31h-4.31ZM6.016 6.016v4.31h4.31v-4.31h-4.31Z" fill="currentColor"></path><path opacity=".75" fill-rule="evenodd" clip-rule="evenodd" d="M3 4.914C3 3.857 3.857 3 4.914 3h6.514c.884 0 1.628.6 1.848 1.414a5.171 5.171 0 0 1 7.31 7.31c.815.22 1.414.964 1.414 1.848v6.514A1.914 1.914 0 0 1 20.086 22H4.914A1.914 1.914 0 0 1 3 20.086V4.914Zm3.016 1.102v4.31h4.31v-4.31h-4.31Zm0 12.968v-4.31h4.31v4.31h-4.31Zm8.658 0v-4.31h4.31v4.31h-4.31Zm0-10.813a2.155 2.155 0 1 1 4.31 0 2.155 2.155 0 0 1-4.31 0Z" fill="currentColor"></path><path opacity=".25" d="M16.829 6.016a2.155 2.155 0 1 0 0 4.31 2.155 2.155 0 0 0 0-4.31Z" fill="currentColor"></path></svg> Spaces</a> </li><li class="hover:text-yellow-700 max-xl:hidden"><a class="group flex items-center px-2 py-0.5 dark:text-gray-300 dark:hover:text-gray-100" href="/posts"><svg class="mr-1.5 text-gray-400 group-hover:text-yellow-500 text-yellow-500!" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" viewBox="0 0 12 12" preserveAspectRatio="xMidYMid meet"><path fill="currentColor" fill-rule="evenodd" d="M3.73 2.4A4.25 4.25 0 1 1 6 10.26H2.17l-.13-.02a.43.43 0 0 1-.3-.43l.01-.06a.43.43 0 0 1 .12-.22l.84-.84A4.26 4.26 0 0 1 3.73 2.4Z" clip-rule="evenodd"></path></svg> Posts</a> </li><li class="hover:text-yellow-700"><a class="group flex items-center px-2 py-0.5 dark:text-gray-300 dark:hover:text-gray-100" href="/docs"><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" class="mr-1.5 text-gray-400 group-hover:text-yellow-500" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 16 16"><path d="m2.28 3.7-.3.16a.67.67 0 0 0-.34.58v8.73l.01.04.02.07.01.04.03.06.02.04.02.03.04.06.05.05.04.04.06.04.06.04.08.04.08.02h.05l.07.02h.11l.04-.01.07-.02.03-.01.07-.03.22-.12a5.33 5.33 0 0 1 5.15.1.67.67 0 0 0 .66 0 5.33 5.33 0 0 1 5.33 0 .67.67 0 0 0 1-.58V4.36a.67.67 0 0 0-.34-.5l-.3-.17v7.78a.63.63 0 0 1-.87.59 4.9 4.9 0 0 0-4.35.35l-.65.39a.29.29 0 0 1-.15.04.29.29 0 0 1-.16-.04l-.65-.4a4.9 4.9 0 0 0-4.34-.34.63.63 0 0 1-.87-.59V3.7Z" fill="currentColor" class="dark:opacity-40"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M8 3.1a5.99 5.99 0 0 0-5.3-.43.66.66 0 0 0-.42.62v8.18c0 .45.46.76.87.59a4.9 4.9 0 0 1 4.34.35l.65.39c.05.03.1.04.16.04.05 0 .1-.01.15-.04l.65-.4a4.9 4.9 0 0 1 4.35-.34.63.63 0 0 0 .86-.59V3.3a.67.67 0 0 0-.41-.62 5.99 5.99 0 0 0-5.3.43l-.3.17L8 3.1Zm.73 1.87a.43.43 0 1 0-.86 0v5.48a.43.43 0 0 0 .86 0V4.97Z" fill="currentColor" class="opacity-40 dark:opacity-100"></path><path d="M8.73 4.97a.43.43 0 1 0-.86 0v5.48a.43.43 0 1 0 .86 0V4.96Z" fill="currentColor" class="dark:opacity-40"></path></svg> Docs</a> </li><li class="hover:text-black dark:hover:text-white"><a class="group flex items-center px-2 py-0.5 dark:text-gray-300 dark:hover:text-gray-100" href="/enterprise"><svg class="mr-1.5 text-gray-400 group-hover:text-black dark:group-hover:text-white" xmlns="http://www.w3.org/2000/svg" fill="none" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 33 27"><path fill="currentColor" fill-rule="evenodd" d="M13.5.7a8.7 8.7 0 0 0-7.7 5.7L1 20.6c-1 3.1.9 5.7 4.1 5.7h15c3.3 0 6.8-2.6 7.8-5.7l4.6-14.2c1-3.1-.8-5.7-4-5.7h-15Zm1.1 5.7L9.8 20.3h9.8l1-3.1h-5.8l.8-2.5h4.8l1.1-3h-4.8l.8-2.3H23l1-3h-9.5Z" clip-rule="evenodd"></path></svg> Enterprise</a> </li> <li><a class="group flex items-center px-2 py-0.5 dark:text-gray-300 dark:hover:text-gray-100" href="/pricing">Pricing </a></li> <li><div class="relative group"> <button class="px-2 py-0.5 hover:text-gray-500 dark:hover:text-gray-600 flex items-center " type="button"> <svg class=" text-gray-500 w-5 group-hover:text-gray-400 dark:text-gray-300 dark:group-hover:text-gray-100" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" viewBox="0 0 32 18" preserveAspectRatio="xMidYMid meet"><path fill-rule="evenodd" clip-rule="evenodd" d="M14.4504 3.30221C14.4504 2.836 14.8284 2.45807 15.2946 2.45807H28.4933C28.9595 2.45807 29.3374 2.836 29.3374 3.30221C29.3374 3.76842 28.9595 4.14635 28.4933 4.14635H15.2946C14.8284 4.14635 14.4504 3.76842 14.4504 3.30221Z" fill="currentColor"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M14.4504 9.00002C14.4504 8.53382 14.8284 8.15588 15.2946 8.15588H28.4933C28.9595 8.15588 29.3374 8.53382 29.3374 9.00002C29.3374 9.46623 28.9595 9.84417 28.4933 9.84417H15.2946C14.8284 9.84417 14.4504 9.46623 14.4504 9.00002Z" fill="currentColor"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M14.4504 14.6978C14.4504 14.2316 14.8284 13.8537 15.2946 13.8537H28.4933C28.9595 13.8537 29.3374 14.2316 29.3374 14.6978C29.3374 15.164 28.9595 15.542 28.4933 15.542H15.2946C14.8284 15.542 14.4504 15.164 14.4504 14.6978Z" fill="currentColor"></path><path fill-rule="evenodd" clip-rule="evenodd" d="M1.94549 6.87377C2.27514 6.54411 2.80962 6.54411 3.13928 6.87377L6.23458 9.96907L9.32988 6.87377C9.65954 6.54411 10.194 6.54411 10.5237 6.87377C10.8533 7.20343 10.8533 7.73791 10.5237 8.06756L6.23458 12.3567L1.94549 8.06756C1.61583 7.73791 1.61583 7.20343 1.94549 6.87377Z" fill="currentColor"></path></svg> </button> </div></li> <li><hr class="h-5 w-0.5 border-none bg-gray-100 dark:bg-gray-800"></li> <li><a class="block cursor-pointer whitespace-nowrap px-2 py-0.5 hover:text-gray-500 dark:text-gray-300 dark:hover:text-gray-100" href="/login">Log In </a></li> <li><a class="whitespace-nowrap rounded-full border border-transparent bg-gray-900 px-3 py-1 leading-none text-white hover:border-black hover:bg-white hover:text-black" href="/join">Sign Up </a></li></ul></nav></div></header></div> <div class="SVELTE_HYDRATER contents" data-target="SSOBanner" data-props="{}"></div> <main class="flex flex-1 flex-col"><div class="relative lg:flex" id="hf-doc-container"><div class="sticky top-0 z-20 self-start"><div class="SVELTE_HYDRATER contents" data-target="SideMenu" data-props="{"chapters":[{"title":"Getting started","isExpanded":true,"sections":[{"title":"TRL","isExpanded":true,"id":"index","url":"/docs/trl/v0.15.2/en/index"},{"title":"Installation","isExpanded":true,"id":"installation","url":"/docs/trl/v0.15.2/en/installation"},{"title":"Quickstart","isExpanded":true,"id":"quickstart","url":"/docs/trl/v0.15.2/en/quickstart"}]},{"title":"Conceptual Guides","isExpanded":true,"sections":[{"title":"Dataset Formats","isExpanded":true,"id":"dataset_formats","url":"/docs/trl/v0.15.2/en/dataset_formats"},{"title":"Training FAQ","isExpanded":true,"id":"how_to_train","url":"/docs/trl/v0.15.2/en/how_to_train"},{"title":"Understanding Logs","isExpanded":true,"id":"logging","url":"/docs/trl/v0.15.2/en/logging"}]},{"title":"How-to guides","isExpanded":true,"sections":[{"title":"Command Line Interface (CLI)","isExpanded":true,"id":"clis","url":"/docs/trl/v0.15.2/en/clis"},{"title":"Customizing the Training","isExpanded":true,"id":"customization","url":"/docs/trl/v0.15.2/en/customization"},{"title":"Reducing Memory Usage","isExpanded":true,"id":"reducing_memory_usage","url":"/docs/trl/v0.15.2/en/reducing_memory_usage"},{"title":"Speeding Up Training","isExpanded":true,"id":"speeding_up_training","url":"/docs/trl/v0.15.2/en/speeding_up_training"},{"title":"Using Trained Models","isExpanded":true,"id":"use_model","url":"/docs/trl/v0.15.2/en/use_model"}]},{"title":"Integrations","isExpanded":true,"sections":[{"title":"DeepSpeed","isExpanded":true,"id":"deepspeed_integration","url":"/docs/trl/v0.15.2/en/deepspeed_integration"},{"title":"Liger Kernel","isExpanded":true,"id":"liger_kernel_integration","url":"/docs/trl/v0.15.2/en/liger_kernel_integration"},{"title":"PEFT","isExpanded":true,"id":"peft_integration","url":"/docs/trl/v0.15.2/en/peft_integration"},{"title":"Unsloth","isExpanded":true,"id":"unsloth_integration","url":"/docs/trl/v0.15.2/en/unsloth_integration"}]},{"title":"Examples","isExpanded":true,"sections":[{"title":"Example Overview","isExpanded":true,"id":"example_overview","url":"/docs/trl/v0.15.2/en/example_overview"},{"title":"Community Tutorials","isExpanded":true,"id":"community_tutorials","url":"/docs/trl/v0.15.2/en/community_tutorials"},{"title":"Sentiment Tuning","isExpanded":true,"id":"sentiment_tuning","url":"/docs/trl/v0.15.2/en/sentiment_tuning"},{"title":"Training StackLlama","isExpanded":true,"id":"using_llama_models","url":"/docs/trl/v0.15.2/en/using_llama_models"},{"title":"Detoxifying a Language Model","isExpanded":true,"id":"detoxifying_a_lm","url":"/docs/trl/v0.15.2/en/detoxifying_a_lm"},{"title":"Learning to Use Tools","isExpanded":true,"id":"learning_tools","url":"/docs/trl/v0.15.2/en/learning_tools"},{"title":"Multi Adapter RLHF","isExpanded":true,"id":"multi_adapter_rl","url":"/docs/trl/v0.15.2/en/multi_adapter_rl"}]},{"title":"API","isExpanded":true,"sections":[{"title":"Trainers","isExpanded":true,"sections":[{"title":"AlignProp","isExpanded":true,"id":"alignprop_trainer","url":"/docs/trl/v0.15.2/en/alignprop_trainer"},{"title":"BCO","isExpanded":true,"id":"bco_trainer","url":"/docs/trl/v0.15.2/en/bco_trainer"},{"title":"CPO","isExpanded":true,"id":"cpo_trainer","url":"/docs/trl/v0.15.2/en/cpo_trainer"},{"title":"DDPO","isExpanded":true,"id":"ddpo_trainer","url":"/docs/trl/v0.15.2/en/ddpo_trainer"},{"title":"DPO","isExpanded":true,"id":"dpo_trainer","url":"/docs/trl/v0.15.2/en/dpo_trainer"},{"title":"Online DPO","isExpanded":true,"id":"online_dpo_trainer","url":"/docs/trl/v0.15.2/en/online_dpo_trainer"},{"title":"GKD","isExpanded":true,"id":"gkd_trainer","url":"/docs/trl/v0.15.2/en/gkd_trainer"},{"title":"GRPO","isExpanded":true,"id":"grpo_trainer","url":"/docs/trl/v0.15.2/en/grpo_trainer"},{"title":"KTO","isExpanded":true,"id":"kto_trainer","url":"/docs/trl/v0.15.2/en/kto_trainer"},{"title":"Nash-MD","isExpanded":true,"id":"nash_md_trainer","url":"/docs/trl/v0.15.2/en/nash_md_trainer"},{"title":"ORPO","isExpanded":true,"id":"orpo_trainer","url":"/docs/trl/v0.15.2/en/orpo_trainer"},{"title":"PPO","isExpanded":true,"id":"ppo_trainer","url":"/docs/trl/v0.15.2/en/ppo_trainer"},{"title":"PRM","isExpanded":true,"id":"prm_trainer","url":"/docs/trl/v0.15.2/en/prm_trainer"},{"title":"Reward","isExpanded":true,"id":"reward_trainer","url":"/docs/trl/v0.15.2/en/reward_trainer"},{"title":"RLOO","isExpanded":true,"id":"rloo_trainer","url":"/docs/trl/v0.15.2/en/rloo_trainer"},{"title":"SFT","isExpanded":true,"id":"sft_trainer","url":"/docs/trl/v0.15.2/en/sft_trainer"},{"title":"Iterative SFT","isExpanded":true,"id":"iterative_sft_trainer","url":"/docs/trl/v0.15.2/en/iterative_sft_trainer"},{"title":"XPO","isExpanded":true,"id":"xpo_trainer","url":"/docs/trl/v0.15.2/en/xpo_trainer"}]},{"title":"Model Classes","isExpanded":true,"id":"models","url":"/docs/trl/v0.15.2/en/models"},{"title":"Best of N Sampling","isExpanded":true,"id":"best_of_n","url":"/docs/trl/v0.15.2/en/best_of_n"},{"title":"Judges","isExpanded":true,"id":"judges","url":"/docs/trl/v0.15.2/en/judges"},{"title":"Callbacks","isExpanded":true,"id":"callbacks","url":"/docs/trl/v0.15.2/en/callbacks"},{"title":"Data Utilities","isExpanded":true,"id":"data_utils","url":"/docs/trl/v0.15.2/en/data_utils"},{"title":"Text Environments","isExpanded":true,"id":"text_environments","url":"/docs/trl/v0.15.2/en/text_environments"},{"title":"Script Utilities","isExpanded":true,"id":"script_utils","url":"/docs/trl/v0.15.2/en/script_utils"}]}],"chapterId":"nash_md_trainer","docType":"docs","isLoggedIn":false,"lang":"en","langs":["en"],"library":"trl","theme":"light","version":"v0.15.2","versions":[{"version":"main"},{"version":"v0.16.0"},{"version":"v0.15.2"},{"version":"v0.15.1"},{"version":"v0.15.0"},{"version":"v0.14.0"},{"version":"v0.13.0"},{"version":"v0.12.2"},{"version":"v0.12.1"},{"version":"v0.12.0"},{"version":"v0.11.4"},{"version":"v0.11.3"},{"version":"v0.11.2"},{"version":"v0.11.1"},{"version":"v0.11.0"},{"version":"v0.10.1"},{"version":"v0.9.6"},{"version":"v0.9.4"},{"version":"v0.9.3"},{"version":"v0.9.2"},{"version":"v0.8.6"},{"version":"v0.8.5"},{"version":"v0.8.4"},{"version":"v0.8.3"},{"version":"v0.8.2"},{"version":"v0.8.1"},{"version":"v0.8.0"},{"version":"v0.7.11"},{"version":"v0.7.10"},{"version":"v0.7.9"},{"version":"v0.7.8"},{"version":"v0.7.4"},{"version":"v0.7.2"},{"version":"v0.7.1"},{"version":"v0.7.0"},{"version":"v0.6.0"},{"version":"v0.5.0"},{"version":"v0.4.7"},{"version":"v0.4.6"},{"version":"v0.4.5"},{"version":"v0.4.2"},{"version":"v0.4.1"},{"version":"v0.4.0"},{"version":"v0.3.1"},{"version":"v0.3.0"},{"version":"v0.2.1"},{"version":"v0.2.0"},{"version":"v0.1.1"}],"title":"Nash-MD Trainer"}"> <div class="z-2 w-full flex-none lg:flex lg:h-dvh lg:w-[270px] lg:flex-col 2xl:w-[300px] false"><div class="shadow-alternate flex h-auto w-full items-center rounded-b-xl border-b bg-white py-2 text-lg leading-tight lg:hidden"> <div class="flex flex-1 cursor-pointer flex-col justify-center self-stretch pl-6"><p class="text-sm text-gray-400 first-letter:capitalize">TRL documentation </p> <div class="mr-2 flex items-center"><p class="font-semibold">Nash-MD Trainer</p> <svg class="text-xl false" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24"><path d="M16.293 9.293L12 13.586L7.707 9.293l-1.414 1.414L12 16.414l5.707-5.707z" fill="currentColor"></path></svg></div></div> <button class="hover:shadow-alternate group ml-auto mr-6 inline-flex flex-none cursor-pointer rounded-xl border p-2"><svg class="text-gray-500 group-hover:text-gray-700" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M30 28.59L22.45 21A11 11 0 1 0 21 22.45L28.59 30zM5 14a9 9 0 1 1 9 9a9 9 0 0 1-9-9z" fill="currentColor"></path></svg></button></div> <div class="bg-linear-to-r hidden flex-col justify-between border-b border-r bg-white p-4 lg:flex from-blue-50 to-white dark:from-gray-900 dark:to-gray-950"><div class="group relative mb-2 flex min-w-[50%] items-center self-start text-lg font-bold leading-tight first-letter:capitalize"><div class="mr-1.5 h-1.5 w-1.5 rounded-full bg-blue-500 flex-none"></div> <h1>TRL</h1> <svg class="opacity-50 ml-0.5 flex-none group-hover:opacity-100" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24"><path d="M16.293 9.293L12 13.586L7.707 9.293l-1.414 1.414L12 16.414l5.707-5.707z" fill="currentColor"></path></svg> <select class="outline-hidden absolute inset-0 border-none bg-white text-base opacity-0"><option value="/docs">🏡 View all docs</option><option value="/docs/optimum-neuron" >AWS Trainium & Inferentia</option><option value="/docs/accelerate" >Accelerate</option><option value="/docs/sagemaker" >Amazon SageMaker</option><option value="https://argilla-io.github.io/argilla/" >Argilla</option><option value="/docs/autotrain" >AutoTrain</option><option value="/docs/bitsandbytes" >Bitsandbytes</option><option value="/docs/chat-ui" >Chat UI</option><option value="/docs/competitions" >Competitions</option><option value="/docs/dataset-viewer" >Dataset viewer</option><option value="/docs/datasets" >Datasets</option><option value="/docs/diffusers" >Diffusers</option><option value="https://distilabel.argilla.io/" >Distilabel</option><option value="/docs/evaluate" >Evaluate</option><option value="https://www.gradio.app/docs/" >Gradio</option><option value="/docs/hub" >Hub</option><option value="/docs/huggingface_hub" >Hub Python Library</option><option value="/docs/hugs" >Hugging Face Generative AI Services (HUGS)</option><option value="/docs/huggingface.js" >Huggingface.js</option><option value="/docs/api-inference" >Inference API (serverless)</option><option value="/docs/inference-endpoints" >Inference Endpoints (dedicated)</option><option value="/docs/leaderboards" >Leaderboards</option><option value="/docs/lighteval" >Lighteval</option><option value="/docs/optimum" >Optimum</option><option value="/docs/peft" >PEFT</option><option value="/docs/safetensors" >Safetensors</option><option value="https://sbert.net/" >Sentence Transformers</option><option value="/docs/trl" selected>TRL</option><option value="/tasks" >Tasks</option><option value="/docs/text-embeddings-inference" >Text Embeddings Inference</option><option value="/docs/text-generation-inference" >Text Generation Inference</option><option value="/docs/tokenizers" >Tokenizers</option><option value="/docs/transformers" >Transformers</option><option value="/docs/transformers.js" >Transformers.js</option><option value="/docs/smolagents" >smolagents</option><option value="/docs/timm" >timm</option></select></div> <button class="shadow-alternate mb-2 flex w-full items-center rounded-full border bg-white px-2 py-1 text-left text-sm text-gray-400 ring-indigo-200 hover:bg-indigo-50 hover:ring-2 dark:border-gray-700 dark:ring-yellow-600 dark:hover:bg-gray-900 dark:hover:text-yellow-500"><svg class="flex-none mr-1.5" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M30 28.59L22.45 21A11 11 0 1 0 21 22.45L28.59 30zM5 14a9 9 0 1 1 9 9a9 9 0 0 1-9-9z" fill="currentColor"></path></svg> <div>Search documentation</div> </button> <div class="flex items-center"> <select class="form-input mt-0! w-20! border! dark:text-gray-400! mr-1 rounded-sm border-gray-200 p-1 text-xs uppercase"><option value="0" >main</option><option value="1" >v0.16.0</option><option value="2" selected>v0.15.2</option><option value="3" >v0.14.0</option><option value="4" >v0.13.0</option><option value="5" >v0.12.2</option><option value="6" >v0.11.4</option><option value="7" >v0.10.1</option><option value="8" >v0.9.6</option><option value="9" >v0.8.6</option><option value="10" >v0.7.11</option><option value="11" >v0.6.0</option><option value="12" >v0.5.0</option><option value="13" >v0.4.7</option><option value="14" >v0.3.1</option><option value="15" >v0.2.1</option><option value="16" >v0.1.1</option></select> <select class="form-input dark:text-gray-400! mr-1 rounded-sm border-gray-200 p-1 text-xs w-12! mt-0! border!"><option value="en" selected>EN</option></select> <div class="relative inline-block "> <button class="rounded-full border border-gray-100 pl-2 py-1 pr-2.5 flex items-center text-sm text-gray-500 bg-white hover:bg-yellow-50 hover:border-yellow-200 dark:hover:bg-gray-800 dark:hover:border-gray-950 dark:border-gray-800 " type="button"> <svg class=" text-yellow-500" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24" fill="currentColor"><path d="M6.05 4.14l-.39-.39a.993.993 0 0 0-1.4 0l-.01.01a.984.984 0 0 0 0 1.4l.39.39c.39.39 1.01.39 1.4 0l.01-.01a.984.984 0 0 0 0-1.4zM3.01 10.5H1.99c-.55 0-.99.44-.99.99v.01c0 .55.44.99.99.99H3c.56.01 1-.43 1-.98v-.01c0-.56-.44-1-.99-1zm9-9.95H12c-.56 0-1 .44-1 .99v.96c0 .55.44.99.99.99H12c.56.01 1-.43 1-.98v-.97c0-.55-.44-.99-.99-.99zm7.74 3.21c-.39-.39-1.02-.39-1.41-.01l-.39.39a.984.984 0 0 0 0 1.4l.01.01c.39.39 1.02.39 1.4 0l.39-.39a.984.984 0 0 0 0-1.4zm-1.81 15.1l.39.39a.996.996 0 1 0 1.41-1.41l-.39-.39a.993.993 0 0 0-1.4 0c-.4.4-.4 1.02-.01 1.41zM20 11.49v.01c0 .55.44.99.99.99H22c.55 0 .99-.44.99-.99v-.01c0-.55-.44-.99-.99-.99h-1.01c-.55 0-.99.44-.99.99zM12 5.5c-3.31 0-6 2.69-6 6s2.69 6 6 6s6-2.69 6-6s-2.69-6-6-6zm-.01 16.95H12c.55 0 .99-.44.99-.99v-.96c0-.55-.44-.99-.99-.99h-.01c-.55 0-.99.44-.99.99v.96c0 .55.44.99.99.99zm-7.74-3.21c.39.39 1.02.39 1.41 0l.39-.39a.993.993 0 0 0 0-1.4l-.01-.01a.996.996 0 0 0-1.41 0l-.39.39c-.38.4-.38 1.02.01 1.41z"></path></svg> </button> </div> <a href="https://github.com/huggingface/trl" class="group ml-auto text-xs text-gray-500 hover:text-gray-700 hover:underline dark:hover:text-gray-300"><svg class="inline-block text-gray-500 group-hover:text-gray-700 dark:group-hover:text-gray-300 mr-1.5 -mt-1 w-4 h-4" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1.03em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 250"><path d="M128.001 0C57.317 0 0 57.307 0 128.001c0 56.554 36.676 104.535 87.535 121.46c6.397 1.185 8.746-2.777 8.746-6.158c0-3.052-.12-13.135-.174-23.83c-35.61 7.742-43.124-15.103-43.124-15.103c-5.823-14.795-14.213-18.73-14.213-18.73c-11.613-7.944.876-7.78.876-7.78c12.853.902 19.621 13.19 19.621 13.19c11.417 19.568 29.945 13.911 37.249 10.64c1.149-8.272 4.466-13.92 8.127-17.116c-28.431-3.236-58.318-14.212-58.318-63.258c0-13.975 5-25.394 13.188-34.358c-1.329-3.224-5.71-16.242 1.24-33.874c0 0 10.749-3.44 35.21 13.121c10.21-2.836 21.16-4.258 32.038-4.307c10.878.049 21.837 1.47 32.066 4.307c24.431-16.56 35.165-13.12 35.165-13.12c6.967 17.63 2.584 30.65 1.255 33.873c8.207 8.964 13.173 20.383 13.173 34.358c0 49.163-29.944 59.988-58.447 63.157c4.591 3.972 8.682 11.762 8.682 23.704c0 17.126-.148 30.91-.148 35.126c0 3.407 2.304 7.398 8.792 6.14C219.37 232.5 256 184.537 256 128.002C256 57.307 198.691 0 128.001 0zm-80.06 182.34c-.282.636-1.283.827-2.194.39c-.929-.417-1.45-1.284-1.15-1.922c.276-.655 1.279-.838 2.205-.399c.93.418 1.46 1.293 1.139 1.931zm6.296 5.618c-.61.566-1.804.303-2.614-.591c-.837-.892-.994-2.086-.375-2.66c.63-.566 1.787-.301 2.626.591c.838.903 1 2.088.363 2.66zm4.32 7.188c-.785.545-2.067.034-2.86-1.104c-.784-1.138-.784-2.503.017-3.05c.795-.547 2.058-.055 2.861 1.075c.782 1.157.782 2.522-.019 3.08zm7.304 8.325c-.701.774-2.196.566-3.29-.49c-1.119-1.032-1.43-2.496-.726-3.27c.71-.776 2.213-.558 3.315.49c1.11 1.03 1.45 2.505.701 3.27zm9.442 2.81c-.31 1.003-1.75 1.459-3.199 1.033c-1.448-.439-2.395-1.613-2.103-2.626c.301-1.01 1.747-1.484 3.207-1.028c1.446.436 2.396 1.602 2.095 2.622zm10.744 1.193c.036 1.055-1.193 1.93-2.715 1.95c-1.53.034-2.769-.82-2.786-1.86c0-1.065 1.202-1.932 2.733-1.958c1.522-.03 2.768.818 2.768 1.868zm10.555-.405c.182 1.03-.875 2.088-2.387 2.37c-1.485.271-2.861-.365-3.05-1.386c-.184-1.056.893-2.114 2.376-2.387c1.514-.263 2.868.356 3.061 1.403z" fill="currentColor"></path></svg> </a></div></div> <nav class="hidden flex-auto lg:flex bottom-0 left-0 w-full flex-col overflow-y-auto border-r px-4 pb-16 pt-3 text-[0.95rem] lg:w-[270px] 2xl:w-[300px]"> <div class="group flex cursor-pointer items-center pl-2 text-[0.8rem] font-semibold uppercase leading-9 hover:text-gray-700 dark:hover:text-gray-300 ml-0"><div class="flex after:absolute after:right-4 after:text-gray-500 group-hover:after:content-['▶'] after:rotate-90 after:transform"><span><span class="inline-block space-x-1 leading-5"><span>Getting started</span> </span></span> </div></div> <div class="flex flex-col"><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/index" id="index">TRL </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/installation" id="installation">Installation </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/quickstart" id="quickstart">Quickstart </a> </div> <div class="group flex cursor-pointer items-center pl-2 text-[0.8rem] font-semibold uppercase leading-9 hover:text-gray-700 dark:hover:text-gray-300 ml-0"><div class="flex after:absolute after:right-4 after:text-gray-500 group-hover:after:content-['▶'] after:rotate-90 after:transform"><span><span class="inline-block space-x-1 leading-5"><span>Conceptual Guides</span> </span></span> </div></div> <div class="flex flex-col"><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/dataset_formats" id="dataset_formats">Dataset Formats </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/how_to_train" id="how_to_train">Training FAQ </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/logging" id="logging">Understanding Logs </a> </div> <div class="group flex cursor-pointer items-center pl-2 text-[0.8rem] font-semibold uppercase leading-9 hover:text-gray-700 dark:hover:text-gray-300 ml-0"><div class="flex after:absolute after:right-4 after:text-gray-500 group-hover:after:content-['▶'] after:rotate-90 after:transform"><span><span class="inline-block space-x-1 leading-5"><span>How-to guides</span> </span></span> </div></div> <div class="flex flex-col"><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/clis" id="clis">Command Line Interface (CLI) </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/customization" id="customization">Customizing the Training </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/reducing_memory_usage" id="reducing_memory_usage">Reducing Memory Usage </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/speeding_up_training" id="speeding_up_training">Speeding Up Training </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/use_model" id="use_model">Using Trained Models </a> </div> <div class="group flex cursor-pointer items-center pl-2 text-[0.8rem] font-semibold uppercase leading-9 hover:text-gray-700 dark:hover:text-gray-300 ml-0"><div class="flex after:absolute after:right-4 after:text-gray-500 group-hover:after:content-['▶'] after:rotate-90 after:transform"><span><span class="inline-block space-x-1 leading-5"><span>Integrations</span> </span></span> </div></div> <div class="flex flex-col"><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/deepspeed_integration" id="deepspeed_integration">DeepSpeed </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/liger_kernel_integration" id="liger_kernel_integration">Liger Kernel </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/peft_integration" id="peft_integration">PEFT </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/unsloth_integration" id="unsloth_integration">Unsloth </a> </div> <div class="group flex cursor-pointer items-center pl-2 text-[0.8rem] font-semibold uppercase leading-9 hover:text-gray-700 dark:hover:text-gray-300 ml-0"><div class="flex after:absolute after:right-4 after:text-gray-500 group-hover:after:content-['▶'] after:rotate-90 after:transform"><span><span class="inline-block space-x-1 leading-5"><span>Examples</span> </span></span> </div></div> <div class="flex flex-col"><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/example_overview" id="example_overview">Example Overview </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/community_tutorials" id="community_tutorials">Community Tutorials </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/sentiment_tuning" id="sentiment_tuning">Sentiment Tuning </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/using_llama_models" id="using_llama_models">Training StackLlama </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/detoxifying_a_lm" id="detoxifying_a_lm">Detoxifying a Language Model </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/learning_tools" id="learning_tools">Learning to Use Tools </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/multi_adapter_rl" id="multi_adapter_rl">Multi Adapter RLHF </a> </div> <div class="group flex cursor-pointer items-center pl-2 text-[0.8rem] font-semibold uppercase leading-9 hover:text-gray-700 dark:hover:text-gray-300 ml-0"><div class="flex after:absolute after:right-4 after:text-gray-500 group-hover:after:content-['▶'] after:rotate-90 after:transform"><span><span class="inline-block space-x-1 leading-5"><span>API</span> </span></span> </div></div> <div class="flex flex-col"> <div class="group flex cursor-pointer items-center pl-2 text-[0.8rem] font-semibold uppercase leading-9 hover:text-gray-700 dark:hover:text-gray-300 ml-2"><div class="flex after:absolute after:right-4 after:text-gray-500 group-hover:after:content-['▶'] after:rotate-90 after:transform"><span><span class="inline-block space-x-1 leading-5"><span>Trainers</span> </span></span> </div></div> <div class="flex flex-col"><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/alignprop_trainer" id="alignprop_trainer">AlignProp </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/bco_trainer" id="bco_trainer">BCO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/cpo_trainer" id="cpo_trainer">CPO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/ddpo_trainer" id="ddpo_trainer">DDPO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/dpo_trainer" id="dpo_trainer">DPO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/online_dpo_trainer" id="online_dpo_trainer">Online DPO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/gkd_trainer" id="gkd_trainer">GKD </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/grpo_trainer" id="grpo_trainer">GRPO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/kto_trainer" id="kto_trainer">KTO </a><a class="bg-linear-to-br rounded-xl from-black to-gray-900 py-1 pl-2 pr-2 text-white first:mt-1 last:mb-4 dark:from-gray-800 dark:to-gray-900 ml-4" href="/docs/trl/v0.15.2/en/nash_md_trainer" id="nash_md_trainer">Nash-MD </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/orpo_trainer" id="orpo_trainer">ORPO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/ppo_trainer" id="ppo_trainer">PPO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/prm_trainer" id="prm_trainer">PRM </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/reward_trainer" id="reward_trainer">Reward </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/rloo_trainer" id="rloo_trainer">RLOO </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/sft_trainer" id="sft_trainer">SFT </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/iterative_sft_trainer" id="iterative_sft_trainer">Iterative SFT </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-4" href="/docs/trl/v0.15.2/en/xpo_trainer" id="xpo_trainer">XPO </a> </div><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/models" id="models">Model Classes </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/best_of_n" id="best_of_n">Best of N Sampling </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/judges" id="judges">Judges </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/callbacks" id="callbacks">Callbacks </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/data_utils" id="data_utils">Data Utilities </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/text_environments" id="text_environments">Text Environments </a><a class="transform py-1 pl-2 pr-2 text-gray-500 first:mt-1 last:mb-4 hover:translate-x-px hover:text-black dark:hover:text-gray-300 ml-2" href="/docs/trl/v0.15.2/en/script_utils" id="script_utils">Script Utilities </a> </div></nav></div></div></div> <div class="z-1 min-w-0 flex-1"><div class="flex justify-center bg-blue-50 px-6 py-1.5 text-xs text-blue-600 dark:bg-blue-900 dark:text-blue-200 sm:text-sm"><div>You are viewing v0.15.2 version. <span>A newer version <a class="underline" href="/docs/trl/v0.16.0/nash_md_trainer">v0.16.0</a> is available.</span></div></div> <div class="px-6 pt-6 md:px-12 md:pb-16 md:pt-16"><div class="max-w-4xl mx-auto mb-10"><div class="bg-linear-to-br relative overflow-hidden rounded-xl from-orange-300/10 px-4 py-5 ring-1 ring-orange-100/70 md:px-6 md:py-8"><img alt="Hugging Face's logo" class="absolute -bottom-6 -right-6 w-28 -rotate-45 md:hidden" src="/front/assets/huggingface_logo-noborder.svg"> <div class="mb-2 text-2xl font-bold dark:text-gray-200 md:mb-0">Join the Hugging Face community</div> <p class="mb-4 text-lg text-gray-400 dark:text-gray-300 md:mb-8">and get access to the augmented documentation experience </p> <div class="mb-8 hidden space-y-4 md:block xl:flex xl:space-x-6 xl:space-y-0"><div class="flex items-center"><div class="bg-linear-to-br mr-3 flex h-9 w-9 flex-none items-center justify-center rounded-lg from-indigo-100 to-indigo-100/20 dark:to-indigo-100"><svg class="text-indigo-400 group-hover:text-indigo-500" style="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24"><path class="uim-quaternary" d="M20.23 7.24L12 12L3.77 7.24a1.98 1.98 0 0 1 .7-.71L11 2.76c.62-.35 1.38-.35 2 0l6.53 3.77c.29.173.531.418.7.71z" opacity=".25" fill="currentColor"></path><path class="uim-tertiary" d="M12 12v9.5a2.09 2.09 0 0 1-.91-.21L4.5 17.48a2.003 2.003 0 0 1-1-1.73v-7.5a2.06 2.06 0 0 1 .27-1.01L12 12z" opacity=".5" fill="currentColor"></path><path class="uim-primary" d="M20.5 8.25v7.5a2.003 2.003 0 0 1-1 1.73l-6.62 3.82c-.275.13-.576.198-.88.2V12l8.23-4.76c.175.308.268.656.27 1.01z" fill="currentColor"></path></svg></div> <div class="text-smd leading-tight text-gray-500 dark:text-gray-300 xl:max-w-[200px] 2xl:text-base">Collaborate on models, datasets and Spaces </div></div> <div class="flex items-center"><div class="bg-linear-to-br mr-3 flex h-9 w-9 flex-none items-center justify-center rounded-lg from-orange-100 to-orange-100/20 dark:to-orange-50"><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" class="text-xl text-yellow-400" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24"><path d="M11 15H6l7-14v8h5l-7 14v-8z" fill="currentColor"></path></svg></div> <div class="text-smd leading-tight text-gray-500 dark:text-gray-300 xl:max-w-[200px] 2xl:text-base">Faster examples with accelerated inference </div></div> <div class="flex items-center"><div class="bg-linear-to-br mr-3 flex h-9 w-9 flex-none items-center justify-center rounded-lg from-gray-500/10 to-gray-500/5"><svg class="text-gray-400" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M14.9804 3C14.9217 3.0002 14.8631 3.00555 14.8054 3.016C11.622 3.58252 8.76073 5.30669 6.77248 7.85653C4.78422 10.4064 3.80955 13.6016 4.03612 16.8271C4.26268 20.0525 5.67447 23.0801 7.99967 25.327C10.3249 27.5738 13.3991 28.8811 16.6304 28.997C16.7944 29.003 16.9584 28.997 17.1204 28.997C19.2193 28.9984 21.2877 28.4943 23.1507 27.5274C25.0137 26.5605 26.6164 25.1592 27.8234 23.442C27.9212 23.294 27.9783 23.1229 27.9889 22.9458C27.9995 22.7687 27.9633 22.592 27.884 22.4333C27.8046 22.2747 27.6848 22.1397 27.5367 22.0421C27.3887 21.9444 27.2175 21.8875 27.0404 21.877C25.0426 21.7017 23.112 21.0693 21.3976 20.0288C19.6832 18.9884 18.231 17.5676 17.1533 15.8764C16.0756 14.1852 15.4011 12.2688 15.1822 10.2754C14.9632 8.28193 15.2055 6.26484 15.8904 4.38C15.9486 4.22913 15.97 4.06652 15.9527 3.90572C15.9354 3.74492 15.8799 3.59059 15.7909 3.45557C15.7019 3.32055 15.5819 3.20877 15.4409 3.12952C15.2999 3.05028 15.142 3.00587 14.9804 3Z" fill="currentColor"></path></svg></div> <div class="text-smd leading-tight text-gray-500 dark:text-gray-300 xl:max-w-[200px] 2xl:text-base">Switch between documentation themes </div></div></div> <div class="flex items-center space-x-2.5"><a href="/join"><button class="bg-linear-to-br shadow-xs rounded-lg bg-white from-gray-100/20 to-gray-200/60 px-5 py-1.5 font-semibold text-gray-700 ring-1 ring-gray-300/60 hover:to-gray-100/70 hover:ring-gray-300/30 active:shadow-inner">Sign Up</button></a> <p class="text-gray-500 dark:text-gray-300">to get started</p></div></div></div> <div class="prose-doc prose relative mx-auto max-w-4xl break-words"> <link href="/docs/trl/v0.15.2/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/entry/start.a67350a3.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/scheduler.d627b047.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/singletons.873c3aa8.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/index.a57a1c33.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/paths.b049ce47.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/entry/app.0e07cdbf.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/index.73c51727.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/nodes/0.fc3919dd.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/each.e59479a4.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/nodes/30.f4a85dfe.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/Tip.a82942ec.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/Docstring.488b7ad2.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/EditOnGithub.859b9ebc.js"> <link rel="modulepreload" href="/docs/trl/v0.15.2/en/_app/immutable/chunks/CodeBlock.b1cdc5f6.js"><meta name="hf:doc:metadata" content="{"title":"Nash-MD Trainer","local":"nash-md-trainer","sections":[{"title":"Overview","local":"overview","sections":[],"depth":2},{"title":"Quick start","local":"quick-start","sections":[],"depth":2},{"title":"Expected dataset type","local":"expected-dataset-type","sections":[],"depth":2},{"title":"Usage tips","local":"usage-tips","sections":[{"title":"Use a reward model","local":"use-a-reward-model","sections":[],"depth":3},{"title":"Encourage EOS token generation","local":"encourage-eos-token-generation","sections":[],"depth":3},{"title":"Logging Completions","local":"logging-completions","sections":[],"depth":3}],"depth":2},{"title":"Example script","local":"example-script","sections":[],"depth":2},{"title":"Logged metrics","local":"logged-metrics","sections":[],"depth":2},{"title":"NashMDTrainer","local":"trl.NashMDTrainer","sections":[],"depth":2},{"title":"NashMDConfig","local":"trl.NashMDConfig","sections":[],"depth":2}],"depth":1}"> <p></p> <h1 class="relative group"><a id="nash-md-trainer" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#nash-md-trainer"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Nash-MD Trainer</span></h1> <p data-svelte-h="svelte-1l4xqlz"><a href="https://huggingface.co/models?other=nash-md,trl" rel="nofollow"><img src="https://img.shields.io/badge/All_models-Nash--MD-blue"></a></p> <h2 class="relative group"><a id="overview" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#overview"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Overview</span></h2> <p data-svelte-h="svelte-13k3dxs">Nash-MD was proposed in the paper <a href="https://huggingface.co/papers/2312.00886" rel="nofollow">Nash Learning from Human Feedback</a> by Rémi Munos, <a href="https://huggingface.co/misovalko" rel="nofollow">Michal Valko</a>, Daniele Calandriello, Mohammad Gheshlaghi Azar, Mark Rowland, Daniel Guo, Yunhao Tang, Matthieu Geist, Thomas Mésnard, and Andrea Michi.</p> <p data-svelte-h="svelte-vfdo9a">The abstract from the paper is the following:</p> <blockquote data-svelte-h="svelte-g2berh"><p>Reinforcement learning from human feedback (RLHF) has emerged as the main paradigm for aligning large language models (LLMs) with human preferences. Typically, RLHF involves the initial step of learning a reward model from human feedback, often expressed as preferences between pairs of text generations produced by a pre-trained LLM. Subsequently, the LLM’s policy is fine-tuned by optimizing it to maximize the reward model through a reinforcement learning algorithm. However, an inherent limitation of current reward models is their inability to fully represent the richness of human preferences and their dependency on the sampling distribution. In this study, we introduce an alternative pipeline for the fine-tuning of LLMs using pairwise human feedback. Our approach entails the initial learning of a preference model, which is conditioned on two inputs given a prompt, followed by the pursuit of a policy that consistently generates responses preferred over those generated by any competing policy, thus defining the Nash equilibrium of this preference model. We term this approach Nash learning from human feedback (NLHF). In the context of a tabular policy representation, we present a novel algorithmic solution, Nash-MD, founded on the principles of mirror descent. This algorithm produces a sequence of policies, with the last iteration converging to the regularized Nash equilibrium. Additionally, we explore parametric representations of policies and introduce gradient descent algorithms for deep-learning architectures. To demonstrate the effectiveness of our approach, we present experimental results involving the fine-tuning of a LLM for a text summarization task. We believe NLHF offers a compelling avenue for preference learning and policy optimization with the potential of advancing the field of aligning LLMs with human preferences.</p></blockquote> <p data-svelte-h="svelte-73d64t">This post-training method was contributed by <a href="https://huggingface.co/kashif" rel="nofollow">Kashif Rasul</a> and <a href="https://huggingface.co/dtiapkin" rel="nofollow">Daniil Tiapkin</a>, <a href="https://huggingface.co/menardprr" rel="nofollow">Pierre Ménard</a>, Daniele Calandriello and <a href="https://huggingface.co/qgallouedec" rel="nofollow">Quentin Gallouédec</a>.</p> <h2 class="relative group"><a id="quick-start" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#quick-start"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Quick start</span></h2> <p data-svelte-h="svelte-apwhsg">This example demonstrates how to train a model using the Nash-MD method. We use the <a href="https://huggingface.co/Qwen/Qwen2-0.5B-Instruct" rel="nofollow">Qwen 0.5B model</a> as the base model and <a href="/docs/trl/v0.15.2/en/judges#trl.PairRMJudge">PairRMJudge</a> as a judge. We use the prompts from the <a href="https://huggingface.co/datasets/openbmb/UltraFeedback" rel="nofollow">UltraFeedback dataset</a>. You can view the prompts in the dataset here:</p> <iframe src="https://huggingface.co/datasets/trl-lib/ultrafeedback-prompt/embed/viewer/default/train?row=0" frameborder="0" width="100%" height="560px"></iframe> <p data-svelte-h="svelte-uqytq6">Below is the script to train the model:</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><span class="hljs-comment"># train_nash_md.py</span> <span class="hljs-keyword">from</span> datasets <span class="hljs-keyword">import</span> load_dataset <span class="hljs-keyword">from</span> trl <span class="hljs-keyword">import</span> NashMDConfig, NashMDTrainer, PairRMJudge <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">"Qwen/Qwen2-0.5B-Instruct"</span>) tokenizer = AutoTokenizer.from_pretrained(<span class="hljs-string">"Qwen/Qwen2-0.5B-Instruct"</span>) judge = PairRMJudge() train_dataset = load_dataset(<span class="hljs-string">"trl-lib/ultrafeedback-prompt"</span>, split=<span class="hljs-string">"train"</span>) training_args = NashMDConfig(output_dir=<span class="hljs-string">"Qwen2-0.5B-NashMD"</span>, logging_steps=<span class="hljs-number">10</span>) trainer = NashMDTrainer( model=model, judge=judge, args=training_args, processing_class=tokenizer, train_dataset=train_dataset ) trainer.train()</pre></div> <p data-svelte-h="svelte-15hino8">Execute the script using the following command:</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class="">accelerate launch train_nash_md.py</pre></div> <p data-svelte-h="svelte-1mf6g0b">Distributed across 8 GPUs, the training takes approximately 3 hours.</p> <p data-svelte-h="svelte-1yv3ryk">To see how the <a href="https://huggingface.co/trl-lib/Qwen2-0.5B-NashMD" rel="nofollow">trained model</a> performs, you can use the <a href="clis#chat-interface">TRL Chat CLI</a>.</p> <pre data-svelte-h="svelte-1oagwxw"><code>$ trl chat --model_name_or_path trl-lib/Qwen2-0.5B-NashMD <strong><span style="color: red;"><quentin_gallouedec>:</span></strong> What is the best programming language? <strong><span style="color: blue;"><trl-lib/Qwen2-0.5B-NashMD>:</span></strong> The best programming language depends on personal preference, the complexity of the project, and the specific requirements of the task. Some programming languages that are often recommended include Python, Java, and JavaScript, and there are many other languages to choose from depending on individual needs. </code></pre> <h2 class="relative group"><a id="expected-dataset-type" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#expected-dataset-type"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Expected dataset type</span></h2> <p data-svelte-h="svelte-1cdm9ft">Nash-MD requires a <a href="dataset_formats#prompt-only">prompt-only dataset</a>. The <a href="/docs/trl/v0.15.2/en/nash_md_trainer#trl.NashMDTrainer">NashMDTrainer</a> supports both <a href="dataset_formats#conversational">conversational</a> and <a href="dataset_formats#standard">standard</a> dataset format. When provided with a conversational dataset, the trainer will automatically apply the chat template to the dataset.</p> <h2 class="relative group"><a id="usage-tips" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#usage-tips"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Usage tips</span></h2> <h3 class="relative group"><a id="use-a-reward-model" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#use-a-reward-model"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Use a reward model</span></h3> <p data-svelte-h="svelte-1g055pk">Instead of a judge, you can chose to use a reward model — see <a href="https://huggingface.co/spaces/allenai/reward-bench" rel="nofollow">Reward Bench</a> for a leaderboard of public models you can use. Below is a code example showing how to replace a judge with the <a href="https://huggingface.co/trl-lib/Qwen2-0.5B-Reward" rel="nofollow">trl-lib/Qwen2-0.5B-Reward</a> model:</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><span class="hljs-deletion">- from trl import PairRMJudge</span> <span class="hljs-addition">+ from transformers import AutoModelForSequenceClassification</span> <span class="hljs-deletion">- judge = PairRMJudge()</span> <span class="hljs-addition">+ reward_model = AutoModelForSequenceClassification.from_pretrained("trl-lib/Qwen2-0.5B-Reward", num_labels=1)</span> trainer = NashMDTrainer( ... <span class="hljs-deletion">- judge=judge,</span> <span class="hljs-addition">+ reward_model=reward_model,</span> )</pre></div> <div class="course-tip course-tip-orange bg-gradient-to-br dark:bg-gradient-to-r before:border-orange-500 dark:before:border-orange-800 from-orange-50 dark:from-gray-900 to-white dark:to-gray-950 border border-orange-50 text-orange-700 dark:text-gray-400"><p data-svelte-h="svelte-y3ex9p">Make sure that the SFT model and reward model use the <em>same</em> chat template and the same tokenizer. Otherwise, you may find the model completions are scored incorrectly during training.</p></div> <h3 class="relative group"><a id="encourage-eos-token-generation" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#encourage-eos-token-generation"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Encourage EOS token generation</span></h3> <p data-svelte-h="svelte-1haouqj">We may want the model to generate completions within a given length. During training, the model will generate completions up to the maximum length specified in the <code>max_new_tokens</code> argument of <a href="/docs/trl/v0.15.2/en/nash_md_trainer#trl.NashMDConfig">NashMDConfig</a>. If you want to penalize the model for not generating an EOS token before reaching the maximum length, you can use the <code>missing_eos_penalty</code> argument of <a href="/docs/trl/v0.15.2/en/nash_md_trainer#trl.NashMDConfig">NashMDConfig</a>:</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class="">training_args = NashMDConfig(..., max_new_tokens=<span class="hljs-number">128</span>, missing_eos_penalty=<span class="hljs-number">1.0</span>)</pre></div> <h3 class="relative group"><a id="logging-completions" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#logging-completions"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Logging Completions</span></h3> <p data-svelte-h="svelte-l8xv6r">To better understand your model’s behavior during training, you can log sample completions periodically using the <a href="/docs/trl/v0.15.2/en/callbacks#trl.LogCompletionsCallback">LogCompletionsCallback</a>.</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class="">trainer = NashMDTrainer(..., eval_dataset=eval_dataset) completions_callback = LogCompletionsCallback(trainer, num_prompts=<span class="hljs-number">8</span>) trainer.add_callback(completions_callback)</pre></div> <p data-svelte-h="svelte-1hvqa3l">This callback logs the model’s generated completions directly to Weights & Biases.</p> <p data-svelte-h="svelte-1hcdz34"><img src="https://huggingface.co/datasets/trl-lib/documentation-images/resolve/main/wandb_completions.png" alt="Logged Completions"></p> <h2 class="relative group"><a id="example-script" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#example-script"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Example script</span></h2> <p data-svelte-h="svelte-1vqf8nm">We provide an example script to train a model using the Nash-MD method. The script is available in <a href="https://github.com/huggingface/trl/blob/main/examples/scripts/nash_md.py" rel="nofollow"><code>examples/scripts/nash_md.py</code></a></p> <p data-svelte-h="svelte-1v4wk0w">To test the online DPO script with the <a href="https://huggingface.co/trl-lib/Qwen/Qwen2.5-0.5B-Instruct" rel="nofollow">Qwen2.5 0.5B model</a> on the <a href="https://huggingface.co/datasets/openbmb/UltraFeedback" rel="nofollow">UltraFeedback dataset</a>, run the following command:</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class="">python examples/scripts/nash_md.py \ --model_name_or_path Qwen/Qwen2.5-0.5B-Instruct \ --judge pair_rm \ --dataset_name trl-lib/ultrafeedback-prompt \ --learning_rate 5.0e-7 \ --logging_steps 25 \ --output_dir Qwen2.5-0.5B-NashMD-PairRM \ --warmup_ratio 0.1 \ --push_to_hub</pre></div> <h2 class="relative group"><a id="logged-metrics" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#logged-metrics"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>Logged metrics</span></h2> <p data-svelte-h="svelte-1rzdw7s">The logged metrics are as follows:</p> <ul data-svelte-h="svelte-isv9gr"><li><code>loss/kl</code>: The mean KL divergence between the model and reference data.</li> <li><code>objective/entropy</code>: The mean entropy of the model and reference data.</li> <li><code>loss/score</code>: The mean reinforce score loss.</li> <li><code>rewards/chosen</code>: The mean scores (according to the reward model) of the model completions.</li> <li><code>rewards/rejected</code>: The mean scores (according to the reward model) of the mixture completions.</li> <li><code>rewards/probabilities</code>: The mean probability (according to the reward model or judge) of the model completions chosen vs the mixture completion.</li> <li><code>rewards/accuracies</code>: The accuracies of the Nash-MD’s implicit reward model.</li> <li><code>rewards/margins</code>: The mean reward margin (according to reward model) between the chosen and mixture completions.</li> <li><code>logps/chosen</code>: The mean log probabilities of the chosen completions.</li> <li><code>logps/rejected</code>: The mean log probabilities of the reference completions.</li> <li><code>val/model_contain_eos_token</code>: The amount of times the model’s output contains the eos token.</li> <li><code>val/ref_contain_eos_token</code>: The amount of times the mixture’s output contains the eos token.</li> <li><code>beta</code>: The parameter that controls the weight of the loss term representing the deviation from the reference model. Typically fixed, but can be made dynamic by passing a list to <a href="/docs/trl/v0.15.2/en/nash_md_trainer#trl.NashMDConfig">NashMDConfig</a>.</li> <li><code>mixture_coef</code>: Logit mixture coefficient for the model and reference model. Typically fixed, but can be made dynamic by passing a list to <a href="/docs/trl/v0.15.2/en/nash_md_trainer#trl.NashMDConfig">NashMDConfig</a>.</li></ul> <h2 class="relative group"><a id="trl.NashMDTrainer" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>NashMDTrainer</span></h2> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"> <div><span class="group flex space-x-1.5 items-center text-gray-800 bg-gradient-to-r rounded-tr-lg -mt-4 -ml-4 pt-3 px-2.5" id="trl.NashMDTrainer"><h3 class="!m-0"><span class="flex-1 break-all md:text-lg bg-gradient-to-r px-2.5 py-1.5 rounded-xl from-indigo-50/70 to-white dark:from-gray-900 dark:to-gray-950 dark:text-indigo-300 text-indigo-700"><svg class="mr-1.5 text-indigo-500 inline-block -mt-0.5" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width=".8em" height=".8em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24"><path class="uim-quaternary" d="M20.23 7.24L12 12L3.77 7.24a1.98 1.98 0 0 1 .7-.71L11 2.76c.62-.35 1.38-.35 2 0l6.53 3.77c.29.173.531.418.7.71z" opacity=".25" fill="currentColor"></path><path class="uim-tertiary" d="M12 12v9.5a2.09 2.09 0 0 1-.91-.21L4.5 17.48a2.003 2.003 0 0 1-1-1.73v-7.5a2.06 2.06 0 0 1 .27-1.01L12 12z" opacity=".5" fill="currentColor"></path><path class="uim-primary" d="M20.5 8.25v7.5a2.003 2.003 0 0 1-1 1.73l-6.62 3.82c-.275.13-.576.198-.88.2V12l8.23-4.76c.175.308.268.656.27 1.01z" fill="currentColor"></path></svg><span class="font-light">class</span> <span class="font-medium">trl.</span><span class="font-semibold">NashMDTrainer</span></span></h3> <a id="trl.NashMDTrainer" class="header-link invisible with-hover:group-hover:visible pr-2" href="#trl.NashMDTrainer"><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></a> <a class="!ml-auto !text-gray-400 !no-underline text-sm flex items-center" href="https://github.com/huggingface/trl/blob/v0.15.2/trl/trainer/nash_md_trainer.py#L62" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span class="hidden md:block mx-0.5 hover:!underline" data-svelte-h="svelte-122apf4">source</span> <span data-svelte-h="svelte-x0xyl0">></span></a></span> <p class="font-mono text-xs md:text-sm !leading-relaxed !my-6"><span data-svelte-h="svelte-8mvn6a">(</span> <span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">model<span class="opacity-60">: typing.Union[transformers.modeling_utils.PreTrainedModel, torch.nn.modules.module.Module] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ref_model<span class="opacity-60">: typing.Union[transformers.modeling_utils.PreTrainedModel, torch.nn.modules.module.Module] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">reward_model<span class="opacity-60">: typing.Union[transformers.modeling_utils.PreTrainedModel, torch.nn.modules.module.Module, NoneType] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">judge<span class="opacity-60">: typing.Optional[trl.trainer.judges.BasePairwiseJudge] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">args<span class="opacity-60">: typing.Optional[trl.trainer.nash_md_config.NashMDConfig] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">data_collator<span class="opacity-60">: typing.Optional[typing.Callable] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">train_dataset<span class="opacity-60">: typing.Union[datasets.arrow_dataset.Dataset, datasets.iterable_dataset.IterableDataset, NoneType] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_dataset<span class="opacity-60">: typing.Union[datasets.arrow_dataset.Dataset, dict[str, datasets.arrow_dataset.Dataset], NoneType] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">processing_class<span class="opacity-60">: typing.Union[transformers.tokenization_utils_base.PreTrainedTokenizerBase, transformers.image_processing_utils.BaseImageProcessor, transformers.feature_extraction_utils.FeatureExtractionMixin, transformers.processing_utils.ProcessorMixin, NoneType] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">peft_config<span class="opacity-60">: typing.Optional[dict] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">compute_metrics<span class="opacity-60">: typing.Optional[typing.Callable[[transformers.trainer_utils.EvalPrediction], dict]] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">callbacks<span class="opacity-60">: typing.Optional[list[transformers.trainer_callback.TrainerCallback]] = None</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">optimizers<span class="opacity-60">: tuple = (None, None)</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">preprocess_logits_for_metrics<span class="opacity-60">: typing.Optional[typing.Callable[[torch.Tensor, torch.Tensor], torch.Tensor]] = None</span></span> </span> <span data-svelte-h="svelte-1jq0pl7">)</span> </p> <div class="!mb-10 relative docstring-details "> <p class="flex items-center font-semibold !mt-2 !mb-2 text-gray-800" data-svelte-h="svelte-lt6pb6">Parameters <span class="flex-auto border-t-2 border-gray-100 dark:border-gray-700 ml-3"></span></p> <ul class="px-2"><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.model" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.model"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>model</strong> (<code>transformers.PreTrainedModel</code>) — The model to train, preferably an <code>AutoModelForCausalLM</code>. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.ref_model" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.ref_model"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>ref_model</strong> (<code>PreTrainedModelWrapper</code>) — Hugging Face transformer model with a casual language modelling head. Used for implicit reward computation and loss. If no reference model is provided, the trainer will create a reference model with the same architecture as the model to be optimized. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.reward_model" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.reward_model"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>reward_model</strong> (<code>transformers.PreTrainedModel</code>) — The reward model to score completions with, preferably an <code>AutoModelForSequenceClassification</code>. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.judge" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.judge"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>judge</strong> (<code>BasePairwiseJudge</code>) — The judge to use for pairwise comparison of model completions. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.args" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.args"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>args</strong> (<code>NashMDConfig</code>) — The NashMD config arguments to use for training. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.data_collator" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.data_collator"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>data_collator</strong> (<code>transformers.DataCollator</code>) — The data collator to use for training. If None is specified, the default data collator (<code>DPODataCollatorWithPadding</code>) will be used which will pad the sequences to the maximum length of the sequences in the batch, given a dataset of paired sequences. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.train_dataset" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.train_dataset"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>train_dataset</strong> (<code>datasets.Dataset</code>) — The dataset to use for training. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.eval_dataset" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.eval_dataset"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>eval_dataset</strong> (<code>datasets.Dataset</code>) — The dataset to use for evaluation. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.processing_class" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.processing_class"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>processing_class</strong> (<code>PreTrainedTokenizerBase</code> or <code>BaseImageProcessor</code> or <code>FeatureExtractionMixin</code> or <code>ProcessorMixin</code>, <em>optional</em>) — Processing class used to process the data. If provided, will be used to automatically process the inputs for the model, and it will be saved along the model to make it easier to rerun an interrupted training or reuse the fine-tuned model. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.peft_config" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.peft_config"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>peft_config</strong> (<code>dict</code>) — The peft config to use for training. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.compute_metrics" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.compute_metrics"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>compute_metrics</strong> (<code>Callable[[EvalPrediction], dict]</code>, <em>optional</em>) — The function to use to compute the metrics. Must take a <code>EvalPrediction</code> and return a dictionary string to metric values. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.callbacks" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.callbacks"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>callbacks</strong> (<code>list[transformers.TrainerCallback]</code>) — The callbacks to use for training. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.optimizers" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.optimizers"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>optimizers</strong> (<code>tuple[torch.optim.Optimizer, torch.optim.lr_scheduler.LambdaLR]</code>) — The optimizer and scheduler to use for training. </span></span> </li><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDTrainer.preprocess_logits_for_metrics" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDTrainer.preprocess_logits_for_metrics"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>preprocess_logits_for_metrics</strong> (<code>Callable[[torch.Tensor, torch.Tensor], torch.Tensor]</code>) — The function to use to preprocess the logits before computing the metrics. </span></span> </li></ul> </div></div> <p data-svelte-h="svelte-14brmim">Initialize NashMDTrainer as a subclass of <a href="/docs/trl/v0.15.2/en/online_dpo_trainer#trl.OnlineDPOConfig">OnlineDPOConfig</a>.</p></div> <h2 class="relative group"><a id="trl.NashMDConfig" class="header-link block pr-1.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDConfig"><span><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span>NashMDConfig</span></h2> <div class="docstring border-l-2 border-t-2 pl-4 pt-3.5 border-gray-100 rounded-tl-xl mb-6 mt-8"> <div><span class="group flex space-x-1.5 items-center text-gray-800 bg-gradient-to-r rounded-tr-lg -mt-4 -ml-4 pt-3 px-2.5" id="trl.NashMDConfig"><h3 class="!m-0"><span class="flex-1 break-all md:text-lg bg-gradient-to-r px-2.5 py-1.5 rounded-xl from-indigo-50/70 to-white dark:from-gray-900 dark:to-gray-950 dark:text-indigo-300 text-indigo-700"><svg class="mr-1.5 text-indigo-500 inline-block -mt-0.5" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" focusable="false" role="img" width=".8em" height=".8em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 24 24"><path class="uim-quaternary" d="M20.23 7.24L12 12L3.77 7.24a1.98 1.98 0 0 1 .7-.71L11 2.76c.62-.35 1.38-.35 2 0l6.53 3.77c.29.173.531.418.7.71z" opacity=".25" fill="currentColor"></path><path class="uim-tertiary" d="M12 12v9.5a2.09 2.09 0 0 1-.91-.21L4.5 17.48a2.003 2.003 0 0 1-1-1.73v-7.5a2.06 2.06 0 0 1 .27-1.01L12 12z" opacity=".5" fill="currentColor"></path><path class="uim-primary" d="M20.5 8.25v7.5a2.003 2.003 0 0 1-1 1.73l-6.62 3.82c-.275.13-.576.198-.88.2V12l8.23-4.76c.175.308.268.656.27 1.01z" fill="currentColor"></path></svg><span class="font-light">class</span> <span class="font-medium">trl.</span><span class="font-semibold">NashMDConfig</span></span></h3> <a id="trl.NashMDConfig" class="header-link invisible with-hover:group-hover:visible pr-2" href="#trl.NashMDConfig"><svg class="" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></a> <a class="!ml-auto !text-gray-400 !no-underline text-sm flex items-center" href="https://github.com/huggingface/trl/blob/v0.15.2/trl/trainer/nash_md_config.py#L20" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span class="hidden md:block mx-0.5 hover:!underline" data-svelte-h="svelte-122apf4">source</span> <span data-svelte-h="svelte-x0xyl0">></span></a></span> <p class="font-mono text-xs md:text-sm !leading-relaxed !my-6"><span data-svelte-h="svelte-8mvn6a">(</span> <span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">output_dir<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">overwrite_output_dir<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">do_train<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">do_eval<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">do_predict<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_strategy<span class="opacity-60">: typing.Union[transformers.trainer_utils.IntervalStrategy, str] = 'no'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">prediction_loss_only<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">per_device_train_batch_size<span class="opacity-60">: int = 8</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">per_device_eval_batch_size<span class="opacity-60">: int = 8</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">per_gpu_train_batch_size<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">per_gpu_eval_batch_size<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">gradient_accumulation_steps<span class="opacity-60">: int = 1</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_accumulation_steps<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_delay<span class="opacity-60">: typing.Optional[float] = 0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">torch_empty_cache_steps<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">learning_rate<span class="opacity-60">: float = 5e-07</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">weight_decay<span class="opacity-60">: float = 0.0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">adam_beta1<span class="opacity-60">: float = 0.9</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">adam_beta2<span class="opacity-60">: float = 0.999</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">adam_epsilon<span class="opacity-60">: float = 1e-08</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">max_grad_norm<span class="opacity-60">: float = 1.0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">num_train_epochs<span class="opacity-60">: float = 3.0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">max_steps<span class="opacity-60">: int = -1</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">lr_scheduler_type<span class="opacity-60">: typing.Union[transformers.trainer_utils.SchedulerType, str] = 'linear'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">lr_scheduler_kwargs<span class="opacity-60">: typing.Union[dict, str, NoneType] = <factory></span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">warmup_ratio<span class="opacity-60">: float = 0.0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">warmup_steps<span class="opacity-60">: int = 0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">log_level<span class="opacity-60">: typing.Optional[str] = 'passive'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">log_level_replica<span class="opacity-60">: typing.Optional[str] = 'warning'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">log_on_each_node<span class="opacity-60">: bool = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">logging_dir<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">logging_strategy<span class="opacity-60">: typing.Union[transformers.trainer_utils.IntervalStrategy, str] = 'steps'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">logging_first_step<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">logging_steps<span class="opacity-60">: float = 500</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">logging_nan_inf_filter<span class="opacity-60">: bool = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">save_strategy<span class="opacity-60">: typing.Union[transformers.trainer_utils.SaveStrategy, str] = 'steps'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">save_steps<span class="opacity-60">: float = 500</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">save_total_limit<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">save_safetensors<span class="opacity-60">: typing.Optional[bool] = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">save_on_each_node<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">save_only_model<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">restore_callback_states_from_checkpoint<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">no_cuda<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">use_cpu<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">use_mps_device<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">seed<span class="opacity-60">: int = 42</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">data_seed<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">jit_mode_eval<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">use_ipex<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">bf16<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fp16<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fp16_opt_level<span class="opacity-60">: str = 'O1'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">half_precision_backend<span class="opacity-60">: str = 'auto'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">bf16_full_eval<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fp16_full_eval<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">tf32<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">local_rank<span class="opacity-60">: int = -1</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ddp_backend<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">tpu_num_cores<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">tpu_metrics_debug<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">debug<span class="opacity-60">: typing.Union[str, typing.List[transformers.debug_utils.DebugOption]] = ''</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">dataloader_drop_last<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_steps<span class="opacity-60">: typing.Optional[float] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">dataloader_num_workers<span class="opacity-60">: int = 0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">dataloader_prefetch_factor<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">past_index<span class="opacity-60">: int = -1</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">run_name<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">disable_tqdm<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">remove_unused_columns<span class="opacity-60">: typing.Optional[bool] = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">label_names<span class="opacity-60">: typing.Optional[typing.List[str]] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">load_best_model_at_end<span class="opacity-60">: typing.Optional[bool] = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">metric_for_best_model<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">greater_is_better<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ignore_data_skip<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fsdp<span class="opacity-60">: typing.Union[typing.List[transformers.trainer_utils.FSDPOption], str, NoneType] = ''</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fsdp_min_num_params<span class="opacity-60">: int = 0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fsdp_config<span class="opacity-60">: typing.Union[dict, str, NoneType] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">tp_size<span class="opacity-60">: typing.Optional[int] = 0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fsdp_transformer_layer_cls_to_wrap<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">accelerator_config<span class="opacity-60">: typing.Union[dict, str, NoneType] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">deepspeed<span class="opacity-60">: typing.Union[dict, str, NoneType] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">label_smoothing_factor<span class="opacity-60">: float = 0.0</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">optim<span class="opacity-60">: typing.Union[transformers.training_args.OptimizerNames, str] = 'adamw_torch'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">optim_args<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">adafactor<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">group_by_length<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">length_column_name<span class="opacity-60">: typing.Optional[str] = 'length'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">report_to<span class="opacity-60">: typing.Union[NoneType, str, typing.List[str]] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ddp_find_unused_parameters<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ddp_bucket_cap_mb<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ddp_broadcast_buffers<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">dataloader_pin_memory<span class="opacity-60">: bool = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">dataloader_persistent_workers<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">skip_memory_metrics<span class="opacity-60">: bool = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">use_legacy_prediction_loop<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">push_to_hub<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">resume_from_checkpoint<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">hub_model_id<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">hub_strategy<span class="opacity-60">: typing.Union[transformers.trainer_utils.HubStrategy, str] = 'every_save'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">hub_token<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">hub_private_repo<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">hub_always_push<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">gradient_checkpointing<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">gradient_checkpointing_kwargs<span class="opacity-60">: typing.Union[dict, str, NoneType] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">include_inputs_for_metrics<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">include_for_metrics<span class="opacity-60">: typing.List[str] = <factory></span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_do_concat_batches<span class="opacity-60">: bool = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">fp16_backend<span class="opacity-60">: str = 'auto'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">evaluation_strategy<span class="opacity-60">: typing.Union[transformers.trainer_utils.IntervalStrategy, str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">push_to_hub_model_id<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">push_to_hub_organization<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">push_to_hub_token<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">mp_parameters<span class="opacity-60">: str = ''</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">auto_find_batch_size<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">full_determinism<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">torchdynamo<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ray_scope<span class="opacity-60">: typing.Optional[str] = 'last'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ddp_timeout<span class="opacity-60">: typing.Optional[int] = 1800</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">torch_compile<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">torch_compile_backend<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">torch_compile_mode<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">dispatch_batches<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">split_batches<span class="opacity-60">: typing.Optional[bool] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">include_tokens_per_second<span class="opacity-60">: typing.Optional[bool] = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">include_num_input_tokens_seen<span class="opacity-60">: typing.Optional[bool] = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">neftune_noise_alpha<span class="opacity-60">: typing.Optional[float] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">optim_target_modules<span class="opacity-60">: typing.Union[NoneType, str, typing.List[str]] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">batch_eval_metrics<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_on_start<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">use_liger_kernel<span class="opacity-60">: typing.Optional[bool] = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">eval_use_gather_object<span class="opacity-60">: typing.Optional[bool] = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">average_tokens_across_devices<span class="opacity-60">: typing.Optional[bool] = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">reward_model_path<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">judge<span class="opacity-60">: typing.Optional[str] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">max_new_tokens<span class="opacity-60">: int = 64</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">max_length<span class="opacity-60">: int = 512</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">temperature<span class="opacity-60">: float = 0.9</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">missing_eos_penalty<span class="opacity-60">: typing.Optional[float] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">beta<span class="opacity-60">: list = <factory></span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">loss_type<span class="opacity-60">: str = 'sigmoid'</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">dataset_num_proc<span class="opacity-60">: typing.Optional[int] = None</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">disable_dropout<span class="opacity-60">: bool = True</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">use_vllm<span class="opacity-60">: bool = False</span></span> </span><span class="comma cursor-default"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">ds3_gather_for_generation<span class="opacity-60">: bool = True</span></span> </span><span class="comma cursor-pointer"><span class="rounded hover:bg-black hover:text-white dark:hover:bg-white dark:hover:text-black">mixture_coef<span class="opacity-60">: list = <factory></span></span> </span> <span data-svelte-h="svelte-1jq0pl7">)</span> </p> <div class="!mb-10 relative docstring-details "> <p class="flex items-center font-semibold !mt-2 !mb-2 text-gray-800" data-svelte-h="svelte-lt6pb6">Parameters <span class="flex-auto border-t-2 border-gray-100 dark:border-gray-700 ml-3"></span></p> <ul class="px-2"><li class="text-base !pl-4 my-3 rounded "><span class="group flex space-x-1.5 items-start"><a id="trl.NashMDConfig.mixture_coef" class="header-link block pr-0.5 text-lg no-hover:hidden with-hover:absolute with-hover:p-1.5 with-hover:opacity-0 with-hover:group-hover:opacity-100 with-hover:right-full" href="#trl.NashMDConfig.mixture_coef"><span><svg class="text-smd" xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg></span></a> <span><strong>mixture_coef</strong> (<code>float</code> or <code>list[float]</code>, <em>optional</em>, defaults to <code>0.5</code>) — Logit mixture coefficient for the model and reference model. If a list of floats is provided then the mixture coefficient is selected for each new epoch and the last coefficient is used for the rest of the epochs. </span></span> </li></ul> </div></div> <p data-svelte-h="svelte-g8nu8h">Configuration class for the <a href="/docs/trl/v0.15.2/en/nash_md_trainer#trl.NashMDTrainer">NashMDTrainer</a>.</p> <p data-svelte-h="svelte-1krfrcl">Subclass of <a href="/docs/trl/v0.15.2/en/online_dpo_trainer#trl.OnlineDPOConfig">OnlineDPOConfig</a> we can use all its arguments and add the following:</p></div> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/trl/blob/main/docs/source/nash_md_trainer.md" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p> <script> { __sveltekit_wrknss = { assets: "/docs/trl/v0.15.2/en", base: "/docs/trl/v0.15.2/en", env: {} }; const element = document.currentScript.parentElement; const data = [null,null]; Promise.all([ import("/docs/trl/v0.15.2/en/_app/immutable/entry/start.a67350a3.js"), import("/docs/trl/v0.15.2/en/_app/immutable/entry/app.0e07cdbf.js") ]).then(([kit, app]) => { kit.start(app, element, { node_ids: [0, 30], data, form: null, error: null }); }); } </script> </div> <div class="SVELTE_HYDRATER contents" data-target="DocFooterNav" data-props="{"classNames":"mx-auto mt-16 flex max-w-4xl items-center pb-8 font-sans font-medium leading-6 xl:mt-32","chapterPrev":{"title":"KTO","isExpanded":true,"id":"kto_trainer","url":"/docs/trl/v0.15.2/en/kto_trainer"},"chapterNext":{"title":"ORPO","isExpanded":true,"id":"orpo_trainer","url":"/docs/trl/v0.15.2/en/orpo_trainer"},"isCourse":false,"isLoggedIn":false}"><div class="mx-auto mt-16 flex max-w-4xl items-center pb-8 font-sans font-medium leading-6 xl:mt-32"><a href="/docs/trl/v0.15.2/en/kto_trainer" class="mr-8 flex transform items-center text-gray-600 transition-all hover:-translate-x-px hover:text-gray-900 dark:hover:text-gray-300"><span class="mr-2 translate-y-px">←</span>KTO</a> <a href="/docs/trl/v0.15.2/en/orpo_trainer" class="ml-auto flex transform items-center text-right text-gray-600 transition-all hover:translate-x-px hover:text-gray-900 dark:hover:text-gray-300">ORPO<span class="ml-2 translate-y-px">→</span></a></div></div></div></div> <div class="sticky top-0 self-start"><div class="SVELTE_HYDRATER contents" data-target="SubSideMenu" data-props="{"chapter":{"title":"Nash-MD Trainer","isExpanded":true,"id":"nash-md-trainer","url":"#nash-md-trainer","sections":[{"title":"Overview","isExpanded":true,"id":"overview","url":"#overview","sections":[]},{"title":"Quick start","isExpanded":true,"id":"quick-start","url":"#quick-start","sections":[]},{"title":"Expected dataset type","isExpanded":true,"id":"expected-dataset-type","url":"#expected-dataset-type","sections":[]},{"title":"Usage tips","isExpanded":true,"id":"usage-tips","url":"#usage-tips","sections":[{"title":"Use a reward model","isExpanded":true,"id":"use-a-reward-model","url":"#use-a-reward-model","sections":[]},{"title":"Encourage EOS token generation","isExpanded":true,"id":"encourage-eos-token-generation","url":"#encourage-eos-token-generation","sections":[]},{"title":"Logging Completions","isExpanded":true,"id":"logging-completions","url":"#logging-completions","sections":[]}]},{"title":"Example script","isExpanded":true,"id":"example-script","url":"#example-script","sections":[]},{"title":"Logged metrics","isExpanded":true,"id":"logged-metrics","url":"#logged-metrics","sections":[]},{"title":"NashMDTrainer","isExpanded":true,"id":"trl.NashMDTrainer","url":"#trl.NashMDTrainer","sections":[]},{"title":"NashMDConfig","isExpanded":true,"id":"trl.NashMDConfig","url":"#trl.NashMDConfig","sections":[]}]}}"> <nav class="hidden h-dvh w-[270px] flex-none flex-col space-y-3 overflow-y-auto break-words border-l pb-16 pl-6 pr-10 pt-24 text-sm lg:flex 2xl:w-[305px]"> <a href="#nash-md-trainer" class=" text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-nash-md-trainer"><wbr>Nash-M<wbr>D <wbr>Trainer</a> <a href="#overview" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-overview"><wbr>Overview</a> <a href="#quick-start" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-quick-start"><wbr>Quick start</a> <a href="#expected-dataset-type" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-expected-dataset-type"><wbr>Expected dataset type</a> <a href="#usage-tips" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-usage-tips"><wbr>Usage tips</a> <a href="#use-a-reward-model" class="pl-8 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-use-a-reward-model"><wbr>Use a reward model</a> <a href="#encourage-eos-token-generation" class="pl-8 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-encourage-eos-token-generation"><wbr>Encourage EO<wbr>S token generation</a> <a href="#logging-completions" class="pl-8 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-logging-completions"><wbr>Logging <wbr>Completions</a> <a href="#example-script" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-example-script"><wbr>Example script</a> <a href="#logged-metrics" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-logged-metrics"><wbr>Logged metrics</a> <a href="#trl.NashMDTrainer" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-trl.NashMDTrainer"><wbr>NashMD<wbr>Trainer</a> <a href="#trl.NashMDConfig" class="pl-4 text-gray-400 transform hover:translate-x-px hover:text-gray-700 dark:hover:text-gray-300" id="nav-trl.NashMDConfig"><wbr>NashMD<wbr>Config</a> </nav></div></div></div> <div id="doc-footer"></div></main> </div> <script> import("\/front\/build\/kube-859986b\/index.js"); window.moonSha = "kube-859986b\/"; window.__hf_deferred = {}; </script>  <script> if (["hf.co", "huggingface.co"].includes(window.location.hostname)) { const script = document.createElement("script"); script.src = "https://js.stripe.com/v3/"; script.async = true; document.head.appendChild(script); } </script> </body> </html>