trl/pr_2482/en/liger_kernel_integration.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Liger Kernel Integration&quot;,&quot;local&quot;:&quot;liger-kernel-integration&quot;,&quot;sections&quot;:[],&quot;depth&quot;:1}">
		<link href="/docs/trl/pr_2482/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/entry/start.b1dce06a.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/scheduler.d627b047.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/singletons.127162ec.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/index.a57a1c33.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/paths.f74687d9.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/entry/app.d588b1ef.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/index.73c51727.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/nodes/0.6905f3b1.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/each.e59479a4.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/nodes/26.86c58b9c.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/Tip.a82942ec.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/CodeBlock.5f78c87f.js">
		<link rel="modulepreload" href="/docs/trl/pr_2482/en/_app/immutable/chunks/getInferenceSnippets.80273291.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Liger Kernel Integration&quot;,&quot;local&quot;:&quot;liger-kernel-integration&quot;,&quot;sections&quot;:[],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END -->      <p></p>  <h1 class="relative group"><a id="liger-kernel-integration" 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="#liger-kernel-integration"><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>Liger Kernel Integration</span></h1>  <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-txx63b">Section under construction. Feel free to contribute!</p></div> <p data-svelte-h="svelte-1xp5liz"><a href="https://github.com/linkedin/Liger-Kernel" rel="nofollow">Liger Kernel</a> is a collection of Triton kernels designed specifically for LLM training. It can effectively increase multi-GPU training throughput by 20% and reduce memory usage by 60%. That way, we can <strong>4x</strong> our context length, as described in the benchmark below. They have implemented Hugging Face compatible <code>RMSNorm</code>, <code>RoPE</code>, <code>SwiGLU</code>, <code>CrossEntropy</code>, <code>FusedLinearCrossEntropy</code>, with more to come. The kernel works out of the box with <a href="https://github.com/Dao-AILab/flash-attention" rel="nofollow">FlashAttention</a>, <a href="https://pytorch.org/tutorials/intermediate/FSDP_tutorial.html" rel="nofollow">PyTorch FSDP</a>, and <a href="https://github.com/microsoft/DeepSpeed" rel="nofollow">Microsoft DeepSpeed</a>.</p> <p data-svelte-h="svelte-m20qmf">With this memory reduction, you can potentially turn off <code>cpu_offloading</code> or gradient checkpointing to further boost the performance.</p> <table data-svelte-h="svelte-1jpb79"><thead><tr><th>Speed Up</th> <th>Memory Reduction</th></tr></thead> <tbody><tr><td><img src="https://raw.githubusercontent.com/linkedin/Liger-Kernel/main/docs/images/e2e-tps.png" alt="Speed up"></td> <td><img src="https://raw.githubusercontent.com/linkedin/Liger-Kernel/main/docs/images/e2e-memory.png" alt="Memory"></td></tr></tbody></table> <ol data-svelte-h="svelte-1ho27rr"><li>To use Liger-Kernel in <a href="/docs/trl/pr_2482/en/sft_trainer#trl.SFTTrainer">SFTTrainer</a>, first install it by:</li></ol> <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=""><!-- HTML_TAG_START -->pip install liger-kernel<!-- HTML_TAG_END --></pre></div> <ol start="2" data-svelte-h="svelte-oenfop"><li>Once installed, set <code>use_liger_kernel</code> in <a href="/docs/trl/pr_2482/en/sft_trainer#trl.SFTConfig">SFTConfig</a>. No other changes are needed!</li></ol> <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=""><!-- HTML_TAG_START -->training_args = SFTConfig(
    use_liger_kernel=<span class="hljs-literal">True</span>,
    ...
)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1h5mxii">To learn more about Liger-Kernel, visit their <a href="https://github.com/linkedin/Liger-Kernel/" rel="nofollow">official repository</a>.</p> <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/liger_kernel_integration.md" target="_blank"><span data-svelte-h="svelte-1kd6by1">&lt;</span> <span data-svelte-h="svelte-x0xyl0">&gt;</span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a>  <p></p> 
			
			<script>
				{
					__sveltekit_1rhatiq = {
						assets: "/docs/trl/pr_2482/en",
						base: "/docs/trl/pr_2482/en",
						env: {}
					};

					const element = document.currentScript.parentElement;

					const data = [null,null];

					Promise.all([
						import("/docs/trl/pr_2482/en/_app/immutable/entry/start.b1dce06a.js"),
						import("/docs/trl/pr_2482/en/_app/immutable/entry/app.d588b1ef.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 26],
							data,
							form: null,
							error: null
						});
					});
				}
			</script>