Buckets:

hf-doc-build
/

doc-dev

Files
xet
hf-doc-build/doc-dev / trl /pr_7115 /en /index.html
HuggingFaceDocBuilder
17 days ago
download
raw
36.3 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;TRL - Transformers Reinforcement Learning&quot;,&quot;local&quot;:&quot;trl---transformers-reinforcement-learning&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;🎉 What’s New&quot;,&quot;local&quot;:&quot;-whats-new&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Taxonomy&quot;,&quot;local&quot;:&quot;taxonomy&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Online methods&quot;,&quot;local&quot;:&quot;online-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Reward modeling&quot;,&quot;local&quot;:&quot;reward-modeling&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Offline methods&quot;,&quot;local&quot;:&quot;offline-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Knowledge distillation&quot;,&quot;local&quot;:&quot;knowledge-distillation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Experimental&quot;,&quot;local&quot;:&quot;experimental&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Online methods&quot;,&quot;local&quot;:&quot;online-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Offline methods&quot;,&quot;local&quot;:&quot;offline-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Knowledge distillation&quot;,&quot;local&quot;:&quot;knowledge-distillation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4}],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Learn&quot;,&quot;local&quot;:&quot;learn&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Contents&quot;,&quot;local&quot;:&quot;contents&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Blog posts&quot;,&quot;local&quot;:&quot;blog-posts&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Talks&quot;,&quot;local&quot;:&quot;talks&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/trl/main/en/_app/immutable/entry/start.CF-rTf14.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/chunks/D0Nf7iHP.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/chunks/CEMhVIpn.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/entry/app.DNQxIU8x.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/chunks/NfwA2eFl.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/chunks/CJlUYK3H.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/chunks/IOvv4kHD.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/nodes/0.BUAl1guj.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/nodes/2.TeeAsg0w.js" rel="modulepreload">
		<link href="/docs/trl/main/en/_app/immutable/chunks/BKeKIeed.js" rel="modulepreload">
		<!--1ldsu9f--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;TRL - Transformers Reinforcement Learning&quot;,&quot;local&quot;:&quot;trl---transformers-reinforcement-learning&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;🎉 What’s New&quot;,&quot;local&quot;:&quot;-whats-new&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Taxonomy&quot;,&quot;local&quot;:&quot;taxonomy&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Online methods&quot;,&quot;local&quot;:&quot;online-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Reward modeling&quot;,&quot;local&quot;:&quot;reward-modeling&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Offline methods&quot;,&quot;local&quot;:&quot;offline-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Knowledge distillation&quot;,&quot;local&quot;:&quot;knowledge-distillation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Experimental&quot;,&quot;local&quot;:&quot;experimental&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Online methods&quot;,&quot;local&quot;:&quot;online-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Offline methods&quot;,&quot;local&quot;:&quot;offline-methods&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4},{&quot;title&quot;:&quot;Knowledge distillation&quot;,&quot;local&quot;:&quot;knowledge-distillation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:4}],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Learn&quot;,&quot;local&quot;:&quot;learn&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Contents&quot;,&quot;local&quot;:&quot;contents&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Blog posts&quot;,&quot;local&quot;:&quot;blog-posts&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Talks&quot;,&quot;local&quot;:&quot;talks&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/><!---->
		<link href="/docs/trl/main/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><p></p> <div style="text-align: center"><picture><source media="(prefers-color-scheme: light)" srcset="https://huggingface.co/datasets/trl-lib/documentation-images/resolve/main/trl_banner_light.png"/> <img src="https://huggingface.co/datasets/trl-lib/documentation-images/resolve/main/trl_banner_dark.png"/></picture></div> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: right; margin-left: 10px; display: inline-flex; position: relative; z-index: 10;"><div class="inline-flex rounded-md max-sm:rounded-sm"><button class="inline-flex items-center gap-1 h-7 max-sm:h-7 px-2 max-sm:px-1.5 text-sm font-medium text-gray-800 border border-r-0 rounded-l-md max-sm:rounded-l-sm border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-live="polite"><span class="inline-flex items-center justify-center rounded-md p-0.5 max-sm:p-0 hover:text-gray-800 dark:hover:text-gray-200"><svg class="sm:size-3.5 size-3" 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><!----></span> <span>Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg><!----></button></div> <!--[-1--><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="trl---transformers-reinforcement-learning" 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---transformers-reinforcement-learning"><span><svg 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>TRL - Transformers Reinforcement Learning</span></h1><!--]--><!----> <p>TRL is a full stack library where we provide a set of tools to train transformer language models with methods like Supervised Fine-Tuning (SFT), Group Relative Policy Optimization (GRPO), Direct Preference Optimization (DPO), Reward Modeling, and more.
The library is integrated with 🤗 <a href="https://github.com/huggingface/transformers" rel="nofollow">transformers</a>.</p> <!--[1--><h2 class="relative group"><a id="-whats-new" 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="#-whats-new"><span><svg 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>🎉 What’s New</span></h2><!--]--><!----> <p><strong>⚡ Up to 6.9× longer sequences on the same GPU:</strong> SFT, DPO, KTO, GRPO and RLOO now score tokens with a <a href="reducing_memory_usage">fused LM head</a> <strong>by default</strong>. At 8k tokens that is 52% to 82% less peak VRAM.</p> <!--[1--><h2 class="relative group"><a id="taxonomy" 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="#taxonomy"><span><svg 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>Taxonomy</span></h2><!--]--><!----> <p>Below is an overview of TRL trainers, organized by maturity and method type.</p> <div style="display: flex; justify-content: space-between; width: 100%; gap: 2rem;"><div style="flex: 1; min-width: 0;"><!--[3--><h4 class="relative group"><a id="online-methods" 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="#online-methods"><span><svg 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>Online methods</span></h4><!--]--><!----> <ul><li><a href="grpo_trainer"><code>GRPOTrainer</code></a></li> <li><a href="rloo_trainer"><code>RLOOTrainer</code></a></li></ul> <!--[3--><h4 class="relative group"><a id="reward-modeling" 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="#reward-modeling"><span><svg 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>Reward modeling</span></h4><!--]--><!----> <ul><li><a href="reward_trainer"><code>RewardTrainer</code></a></li></ul></div> <div style="flex: 1; min-width: 0;"><!--[3--><h4 class="relative group"><a id="offline-methods" 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="#offline-methods"><span><svg 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>Offline methods</span></h4><!--]--><!----> <ul><li><a href="sft_trainer"><code>SFTTrainer</code></a></li> <li><a href="dpo_trainer"><code>DPOTrainer</code></a></li> <li><a href="kto_trainer"><code>KTOTrainer</code></a></li></ul> <!--[3--><h4 class="relative group"><a id="knowledge-distillation" 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="#knowledge-distillation"><span><svg 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>Knowledge distillation</span></h4><!--]--><!----> <ul><li><a href="distillation_trainer"><code>DistillationTrainer</code></a></li></ul></div></div> <!--[2--><h3 class="relative group"><a id="experimental" 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="#experimental"><span><svg 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>Experimental</span></h3><!--]--><!----> <div style="display: flex; justify-content: space-between; width: 100%; gap: 2rem;"><div style="flex: 1; min-width: 0;"><!--[3--><h4 class="relative group"><a id="online-methods" 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="#online-methods"><span><svg 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>Online methods</span></h4><!--]--><!----> <ul><li><a href="a2po_trainer"><code>A2POTrainer</code></a></li> <li><a href="async_grpo_trainer"><code>AsyncGRPOTrainer</code></a></li> <li><a href="gmpo"><code>GMPOTrainer</code></a></li> <li><a href="online_dpo_trainer"><code>OnlineDPOTrainer</code></a></li></ul></div> <div style="flex: 1; min-width: 0;"><!--[3--><h4 class="relative group"><a id="offline-methods" 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="#offline-methods"><span><svg 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>Offline methods</span></h4><!--]--><!----> <ul><li><a href="bema_for_reference_model">BEMA for Reference Model</a></li> <li><a href="cpo_trainer"><code>CPOTrainer</code></a></li> <li><a href="orpo_trainer"><code>ORPOTrainer</code></a></li> <li><a href="tpo_trainer"><code>TPOTrainer</code></a></li></ul> <!--[3--><h4 class="relative group"><a id="knowledge-distillation" 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="#knowledge-distillation"><span><svg 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>Knowledge distillation</span></h4><!--]--><!----> <ul><li><a href="async_distillation_trainer"><code>AsyncDistillationTrainer</code></a></li> <li><a href="gkd_trainer"><code>GKDTrainer</code></a></li> <li><a href="gold_trainer"><code>GOLDTrainer</code></a></li> <li><a href="iw_opd_trainer"><code>IWOPDTrainer</code></a></li> <li><a href="sdft_trainer"><code>SDFTTrainer</code></a></li> <li><a href="sdpo_trainer"><code>SDPOTrainer</code></a></li> <li><a href="ssd_trainer"><code>SSDTrainer</code></a></li></ul></div></div> <p>You can also explore TRL-related models, datasets, and demos in the <a href="https://huggingface.co/trl-lib" rel="nofollow">TRL Hugging Face organization</a>.</p> <!--[1--><h2 class="relative group"><a id="learn" 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="#learn"><span><svg 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>Learn</span></h2><!--]--><!----> <p>Learn post-training with TRL and other libraries in 🤗 <a href="https://github.com/huggingface/smol-course" rel="nofollow">smol course</a>.</p> <!--[1--><h2 class="relative group"><a id="contents" 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="#contents"><span><svg 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>Contents</span></h2><!--]--><!----> <p>The documentation is organized into the following sections:</p> <ul><li><strong>Getting Started</strong>: installation and quickstart guide.</li> <li><strong>Conceptual Guides</strong>: dataset formats, training FAQ, and understanding logs.</li> <li><strong>How-to Guides</strong>: reducing memory usage, speeding up training, distributing training, etc.</li> <li><strong>Integrations</strong>: DeepSpeed, PEFT, etc.</li> <li><strong>Examples</strong>: example overview, community tutorials, etc.</li> <li><strong>API</strong>: trainers, utils, etc.</li></ul> <!--[1--><h2 class="relative group"><a id="blog-posts" 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="#blog-posts"><span><svg 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>Blog posts</span></h2><!--]--><!----> <div class="mt-10"><div class="w-full flex flex-col space-y-4 md:space-y-0 md:grid md:grid-cols-2 md:gap-y-4 md:gap-x-5"><a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/delta-weight-sync"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/delta-weight-sync/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on May 27, 2026</p> <p class="text-gray-700">Shipping a Trillion Parameters With a Hub Bucket: Delta Weight Sync in TRL</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/trl-v1"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/trl-v1/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on March 31, 2026</p> <p class="text-gray-700">TRL v1: Post-Training Library That Holds When the Field Invalidates Its Own Assumptions</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/async-rl-training-landscape"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/async-rl-landscape/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on March 10, 2026</p> <p class="text-gray-700">Keep the Tokens Flowing: Lessons from 16 Open-Source RL Libraries</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/ulysses-sp"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/ulysses/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on March 9, 2026</p> <p class="text-gray-700">Ulysses Sequence Parallelism: Training with Million-Token Contexts</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/openenv"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/openenv/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published October 23, 2025</p> <p class="text-gray-700">Building the Open Agent Ecosystem Together: Introducing OpenEnv</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/trl-vlm-alignment"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/trl_vlm/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on August 7, 2025</p> <p class="text-gray-700">Vision Language Model Alignment in TRL ⚡️</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/vllm-colocate"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/vllm-colocate/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on June 3, 2025</p> <p class="text-gray-700">NO GPU left behind: Unlocking Efficiency with Co-located vLLM in TRL</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/liger-grpo"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/liger-grpo/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on May 25, 2025</p> <p class="text-gray-700">🐯 Liger GRPO meets TRL</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/open-r1"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/open-r1/thumbnails.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on January 28, 2025</p> <p class="text-gray-700">Open-R1: a fully open reproduction of DeepSeek-R1</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/dpo_vlm"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/dpo_vlm/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on July 10, 2024</p> <p class="text-gray-700">Preference Optimization for Vision Language Models with TRL</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/putting_rl_back_in_rlhf_with_rloo"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/putting_rl_back_in_rlhf_with_rloo/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on June 12, 2024</p> <p class="text-gray-700">Putting RL back in RLHF</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/unsloth-trl"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/hf_unsloth/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on January 10, 2024</p> <p class="text-gray-700">Make LLM Fine-tuning 2x faster with Unsloth and 🤗 TRL</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/trl-ddpo"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/166_trl_ddpo/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on September 29, 2023</p> <p class="text-gray-700">Finetune Stable Diffusion Models with DDPO via TRL</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/dpo-trl"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/157_dpo_trl/dpo_thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on August 8, 2023</p> <p class="text-gray-700">Fine-tune Llama 2 with DPO</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/stackllama"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/138_stackllama/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on April 5, 2023</p> <p class="text-gray-700">StackLLaMA: A hands-on guide to train LLaMA with RLHF</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/trl-peft"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/133_trl_peft/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on March 9, 2023</p> <p class="text-gray-700">Fine-tuning 20B LLMs with RLHF on a 24GB consumer GPU</p></a> <a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/blog/rlhf"><img src="https://raw.githubusercontent.com/huggingface/blog/main/assets/120_rlhf/thumbnail.png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Published on December 9, 2022</p> <p class="text-gray-700">Illustrating Reinforcement Learning from Human Feedback</p></a></div></div> <!--[1--><h2 class="relative group"><a id="talks" 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="#talks"><span><svg 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>Talks</span></h2><!--]--><!----> <div class="mt-10"><div class="w-full flex flex-col space-y-4 md:space-y-0 md:grid md:grid-cols-2 md:gap-y-4 md:gap-x-5"><a class="!no-underline border dark:border-gray-700 p-5 rounded-lg shadow hover:shadow-lg" href="https://huggingface.co/datasets/trl-lib/documentation-images/resolve/main/Fine%20tuning%20with%20TRL%20(Oct%2025).pdf"><img src="https://huggingface.co/datasets/trl-lib/documentation-images/resolve/main/Fine%20tuning%20with%20TRL%20(Oct%2025).png" alt="thumbnail" class="mt-0"/> <p class="text-gray-500 text-sm">Talk given on October 30, 2025</p> <p class="text-gray-700">Fine tuning with TRL</p></a></div></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/index.md" target="_blank"><svg class="mr-1" 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="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg><!----> <span><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]-->
			
			<script>
				{
					__sveltekit_1ngr50x = {
						base: "/docs/trl/main/en",
						assets: "/docs/trl/main/en"
					};
					const element = document.currentScript.parentElement;
					Promise.all([
						import("/docs/trl/main/en/_app/immutable/entry/start.CF-rTf14.js"),
						import("/docs/trl/main/en/_app/immutable/entry/app.DNQxIU8x.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 2],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 36.3 kB
Xet hash:: 0449f91a759fa297dd6d3f07c73346f375c43d7d36dc9c9f9bd5ffe6f929789e
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.