Buckets:

hf-doc-build
/

doc-dev

Files

xet

hf-doc-build/doc-dev / peft /pr_3270 /en /developer_guides /memory_efficient_training.html

HuggingFaceDocBuilder

21 days ago

download

raw

19.8 kB

	<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Memory Efficient Training","local":"memory-efficient-training","sections":[{"title":"Training memory overview","local":"training-memory-overview","sections":[],"depth":2},{"title":"Choosing the right method","local":"choosing-the-right-method","sections":[],"depth":2},{"title":"Quantization","local":"quantization","sections":[],"depth":2},{"title":"Compilation","local":"compilation","sections":[],"depth":2},{"title":"Gradient Checkpointing","local":"gradient-checkpointing","sections":[],"depth":2},{"title":"Chunked NLL loss","local":"chunked-nll-loss","sections":[],"depth":2}],"depth":1}">
	<link href="/docs/peft/pr_3270/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/entry/start.f28557bc.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/scheduler.78382b47.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/singletons.2ac5b46a.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/index.fadd215c.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/paths.992e82e0.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/entry/app.2273f60f.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/preload-helper.251eab2a.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/index.6dd35eb6.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/nodes/0.aaf7e185.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/each.e59479a4.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/nodes/8.2d021dbe.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/CopyLLMTxtMenu.ddc04658.js">
	<link rel="modulepreload" href="/docs/peft/pr_3270/en/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.3b3e5fc5.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Memory Efficient Training","local":"memory-efficient-training","sections":[{"title":"Training memory overview","local":"training-memory-overview","sections":[],"depth":2},{"title":"Choosing the right method","local":"choosing-the-right-method","sections":[],"depth":2},{"title":"Quantization","local":"quantization","sections":[],"depth":2},{"title":"Compilation","local":"compilation","sections":[],"depth":2},{"title":"Gradient Checkpointing","local":"gradient-checkpointing","sections":[],"depth":2},{"title":"Chunked NLL loss","local":"chunked-nll-loss","sections":[],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <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> </div> <h1 class="relative group"><a id="memory-efficient-training" 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="#memory-efficient-training"><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>Memory Efficient Training</span></h1> <p data-svelte-h="svelte-j7hopd">🤗 PEFT makes fine-tuning parameter efficient, but not automatically memory efficient. This overview collects tips for cutting training memory and links to the detailed guides.</p> <blockquote class="tip" data-svelte-h="svelte-137x5ak"><p>Always consider the basics of choosing a smaller base model, smaller batch size or shorter sequence length to lower your memory usage.</p></blockquote> <h2 class="relative group"><a id="training-memory-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="#training-memory-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>Training memory overview</span></h2> <p data-svelte-h="svelte-1gfey27">Let’s dissect the distribution of training memory so we can reason about potential countermeasures. We will use a large language-model trained with the Adam optimizer as an example. When doing full fine-tuning, we will have the following positions taking up memory:</p> <ol data-svelte-h="svelte-1qj1unn"><li>base model parameters: the memory consumption highly depends on the chosen <code>dtype</code>. The less bits per parameter (16 for float16), the less memory this will take up. A 1B model in float16 (16 bit/2 byte per parameter) will roughly take <code>1e9 × 2 byte = 1.863 GiB</code> of memory.</li> <li>all trainable base model parameters ×3 for gradients (1×) and Adam optimizer states (2×), therefore 5.59GB of memory</li> <li>memory for the intermediate activations between layers, these are hard to predict but mostly depend on the used compute dtype and sequence length / batch size.</li></ol> <p data-svelte-h="svelte-wi8m1h">A smaller base model or a using a smaller compute dtype will reduce all points while using shorter sequences or smaller batches mainly affects gradients and activation memory. Employing PEFT methods will reduce the number of trainable parameters and therefore significantly reduce both gradients and optimizer state, saving a lot of memory.</p> <h2 class="relative group"><a id="choosing-the-right-method" 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="#choosing-the-right-method"><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>Choosing the right method</span></h2> <p data-svelte-h="svelte-15jdsyl">Not every PEFT method is built equally and some formulations are easier to build in a memory efficient manner. If you are on a memory budget it makes sense to check out the <a href="https://huggingface.co/spaces/peft-internal-testing/PEFT-method-comparison" rel="nofollow">PEFT method comparison suite</a> and filter for <strong>maximum</strong> accelerator memory usage. Average accelerator memory usage can be fairly equal across methods but not every method scales equally with activations and sequence length; some methods are more prone to memory spikes than others.</p> <p data-svelte-h="svelte-1eu91mi">Consider <a href="troubleshooting#using-trainable-tokens">using trainable tokens</a> when targeting large layers like language modeling heads or embedding layers to fine-tune specific tokens.</p> <h2 class="relative group"><a id="quantization" 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="#quantization"><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>Quantization</span></h2> <p data-svelte-h="svelte-1h0eeix">Quantization is one of the best ways to reduce memory consumption <em>of the base model</em> and will, depending on the employed quantization, also reduce activation memory. Since the PEFT methods will only take up a small portion of the total number of parameters, PEFT defaults to use a higher precision than the base model. This can also have the effect that adapters can mitigate some of the quality loss incurred by quantization methods. Read the <a href="quantization">PEFT quantization guide</a>.</p> <h2 class="relative group"><a id="compilation" 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="#compilation"><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>Compilation</span></h2> <p data-svelte-h="svelte-io679i">The models we train are composed of operations like matrix multiplications, sums and assignments where each operation produces a new result and, subsequently, needs to take up memory. If those intermediate results are not needed we can fuse these operations and save up on memory. This is just one of many optimizations that <code>torch.compile</code> can do for you, so check out the <a href="torch_compile">PEFT torch.compile guide</a>.</p> <h2 class="relative group"><a id="gradient-checkpointing" 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="#gradient-checkpointing"><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>Gradient Checkpointing</span></h2> <p data-svelte-h="svelte-157rjdv">You can trade memory with computation by only saving every nth gradient between layers and computing the rest on the fly. Check out the <a href="https://huggingface.co/docs/transformers/grad_checkpointing" rel="nofollow">gradient checkpointing</a> documentation of Transformers to learn more.</p> <blockquote class="note" data-svelte-h="svelte-1fyf2kk"><p>When not using Diffusers or Transformers you may need to implement your own gradient checkpointing logic, depending on the training framework that you are using.</p></blockquote> <h2 class="relative group"><a id="chunked-nll-loss" 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="#chunked-nll-loss"><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>Chunked NLL loss</span></h2> <p data-svelte-h="svelte-14mnsnm">Using <a href="https://docs.pytorch.org/docs/stable/generated/torch.nn.NLLLoss.html" rel="nofollow"><code>NLLLoss</code></a> is very common when training language models (or classification tasks). You allocate a matrix of size <code>batch × sequence × vocabulary</code>. With particularly long sequences or vocabularies this can get expensive fast.</p> <p data-svelte-h="svelte-dyev6n">When using <a href="https://huggingface.co/docs/trl" rel="nofollow">TRL</a> you can either use the <a href="https://huggingface.co/docs/trl/liger_kernel_integration" rel="nofollow">Liger kernel integration</a> or use <a href="https://huggingface.co/docs/trl/v1.5.1/en/reducing_memory_usage#chunked-cross-entropy-for-reducing-peak-memory-usage" rel="nofollow">Chunked NLLLoss</a>. The latter will split the sequence in chunks of size 256 to keep the maximum memory consumption constant.</p> <p data-svelte-h="svelte-a83hp6"><img src="https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/peft/chunked_nll.png" alt="NLL vs. Chunked NLL comparison"></p> <p data-svelte-h="svelte-1yrojuy">In case the default chunk size is not optimal for your setting, look in the <a href="https://github.com/huggingface/trl/pull/5575" rel="nofollow">original TRL PR</a> for more information on how to tune the chunk size.</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/peft/blob/main/docs/source/developer_guides/memory_efficient_training.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 data-svelte-h="svelte-zjs2n5"><span class="underline">Update</span> on GitHub</span></a> <p></p>

	<script>
	{
	__sveltekit_1xnt4vv = {
	assets: "/docs/peft/pr_3270/en",
	base: "/docs/peft/pr_3270/en",
	env: {}
	};

	const element = document.currentScript.parentElement;

	const data = [null,null];

	Promise.all([
	import("/docs/peft/pr_3270/en/_app/immutable/entry/start.f28557bc.js"),
	import("/docs/peft/pr_3270/en/_app/immutable/entry/app.2273f60f.js")
	]).then(([kit, app]) => {
	kit.start(app, element, {
	node_ids: [0, 8],
	data,
	form: null,
	error: null
	});
	});
	}
	</script>

Xet Storage Details

Size:: 19.8 kB
Xet hash:: 9818eb878a20dd346d9d83c922e1f087a0dfb265927d381db4f3acd203f2551d

Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.