Hugging Face's logo

Buckets:
hf-doc-build
/
doc

Files
xet
hf-doc-build/doc / transformers /main /en /deepspeed.html
HuggingFaceDocBuilder's picture
HuggingFaceDocBuilder
download
raw
53.9 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;DeepSpeed ZeRO&quot;,&quot;local&quot;:&quot;deepspeed-zero&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Installation&quot;,&quot;local&quot;:&quot;installation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Configure&quot;,&quot;local&quot;:&quot;configure&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;ZeRO stages&quot;,&quot;local&quot;:&quot;zero-stages&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Checkpoints&quot;,&quot;local&quot;:&quot;checkpoints&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Next steps&quot;,&quot;local&quot;:&quot;next-steps&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
 <link href="/docs/transformers/main/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/entry/start.3d6cca8a.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/scheduler.31fdf58d.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/singletons.6af0ff6e.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/index.252883d5.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/paths.299a376b.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/entry/app.b6ccab0d.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/preload-helper.c438fa0a.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/index.2f76fdf0.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/nodes/0.f2629ed0.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/each.e59479a4.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/nodes/43.e045c0fb.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/CopyLLMTxtMenu.ad38f6ea.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.fd2f7a8a.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/IconCopy.ac192424.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/CodeBlock.e52df5d6.js">
 <link rel="modulepreload" href="/docs/transformers/main/en/_app/immutable/chunks/HfOption.fb051768.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;DeepSpeed ZeRO&quot;,&quot;local&quot;:&quot;deepspeed-zero&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Installation&quot;,&quot;local&quot;:&quot;installation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Configure&quot;,&quot;local&quot;:&quot;configure&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;ZeRO stages&quot;,&quot;local&quot;:&quot;zero-stages&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Checkpoints&quot;,&quot;local&quot;:&quot;checkpoints&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Next steps&quot;,&quot;local&quot;:&quot;next-steps&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;: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="deepspeed-zero" 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="#deepspeed-zero"><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>DeepSpeed ZeRO</span></h1> <p data-svelte-h="svelte-ek1jb"><a href="https://www.deepspeed.ai/" rel="nofollow">DeepSpeed</a> ZeRO (Zero Redundancy Optimizer) eliminates memory redundancy across distributed training by sharding optimizer states, gradients, and parameters across GPUs. ZeRO has three stages, each sharding more state than the last. DeepSpeed also supports offloading to CPU or NVMe memory for further savings. Every additional stage and offload level reduces peak memory, at the cost of more inter-GPU communication.</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="language-text "><!-- HTML_TAG_START --> params grads opt states
 ┌──────────┐ ┌──────────┐ ┌──────────┐
ZeRO-1 │██████████│ │██████████│ │███░░░░░░░│ GPU 0
 │██████████│ │██████████│ │░░░███░░░░│ GPU 1
 │██████████│ │██████████│ │░░░░░░████│ GPU 2
 └──────────┘ └──────────┘ └──────────┘
 ┌──────────┐ ┌──────────┐ ┌──────────┐
ZeRO-2 │██████████│ │███░░░░░░░│ │███░░░░░░░│ GPU 0
 │██████████│ │░░░███░░░░│ │░░░███░░░░│ GPU 1
 │██████████│ │░░░░░░████│ │░░░░░░████│ GPU 2
 └──────────┘ └──────────┘ └──────────┘
 ┌──────────┐ ┌──────────┐ ┌──────────┐
ZeRO-3 │███░░░░░░░│ │███░░░░░░░│ │███░░░░░░░│ GPU 0
 │░░░███░░░░│ │░░░███░░░░│ │░░░███░░░░│ GPU 1
 │░░░░░░████│ │░░░░░░████│ │░░░░░░████│ GPU 2
 └──────────┘ └──────────┘ └──────────┘
 █ resident ░ held on another GPU<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-13oef8o">ZeRO-2 shards gradients and optimizer states with lower communication overhead than ZeRO-3. Use ZeRO-3 only when your model doesn’t fit across GPUs with ZeRO-2.</p> <h2 class="relative group"><a id="installation" 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="#installation"><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>Installation</span></h2> <p data-svelte-h="svelte-wxzb03">Install DeepSpeed from PyPI, or install Transformers with the <code>deepspeed</code> extra.</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="language-shell "><!-- HTML_TAG_START -->pip install deepspeed
pip install transformers[deepspeed]<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1wy2982">If you run into CUDA-related install errors, check the <a href="./debugging#deepspeed-cuda">DeepSpeed CUDA</a> docs. <a href="https://www.deepspeed.ai/tutorials/advanced-install/#install-deepspeed-from-source" rel="nofollow">Installing from source</a> is the more reliable option because it matches your exact hardware and includes features not yet available in the PyPI release.</p> <h2 class="relative group"><a id="configure" 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="#configure"><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>Configure</span></h2> <p data-svelte-h="svelte-1r93kyb"><a href="/docs/transformers/main/en/main_classes/trainer#transformers.Trainer">Trainer</a> integrates DeepSpeed through the <a href="https://huggingface.co/docs/transformers/main_classes/trainer#transformers.TrainingArguments.deepspeed" rel="nofollow">TrainingArguments.deepspeed</a> argument, which accepts a JSON config file. Alternatively, use an <a href="./accelerate#accelerate-config-file">Accelerate config file</a> instead of <a href="/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a>.</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">TrainingArguments </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Accelerate config file </div></div> <div class="language-select"><p data-svelte-h="svelte-1rcewvt">Use <code>&quot;auto&quot;</code> in your config for values you want DeepSpeed to fill from <a href="/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a>. If you want to explicitly specify a value, make sure you use the <em>same</em> value for both the DeepSpeed argument and <a href="/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a>.</p> <blockquote class="note" data-svelte-h="svelte-g9ivr2"><p>See the <a href="https://www.deepspeed.ai/docs/config-json/" rel="nofollow">DeepSpeed Configuration JSON</a> reference for a complete list of DeepSpeed config options.</p></blockquote> <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="language-json "><!-- HTML_TAG_START --><span class="hljs-attr">&quot;train_micro_batch_size_per_gpu&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-comment">// ← per_device_train_batch_size in TrainingArguments</span>
<span class="hljs-attr">&quot;gradient_accumulation_steps&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-comment">// ← gradient_accumulation_steps in TrainingArguments</span>
<span class="hljs-attr">&quot;optimizer.params.lr&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-comment">// ← learning_rate in TrainingArguments</span>
<span class="hljs-attr">&quot;fp16.enabled&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-comment">// ← fp16 flag in TrainingArguments</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1cegpb1">Pass the config to the <code>deepspeed</code> argument.</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="language-py "><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> TrainingArguments
args = TrainingArguments(
 deepspeed=<span class="hljs-string">&quot;path/to/deepspeed_config.json&quot;</span>,
 ...
)<!-- HTML_TAG_END --></pre></div> <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="language-cli "><!-- HTML_TAG_START --><span class="hljs-comment"># DeepSpeed launcher</span>
deepspeed <span class="hljs-comment">--num_gpus 4 train.py</span>
<span class="hljs-comment"># torchrun</span>
torchrun <span class="hljs-comment">--nproc_per_node 4 train.py</span>
<span class="hljs-comment"># Accelerate</span>
accelerate <span class="hljs-built_in">launch</span> <span class="hljs-comment">--num_processes 4 train.py</span><!-- HTML_TAG_END --></pre></div> </div> <h2 class="relative group"><a id="zero-stages" 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="#zero-stages"><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>ZeRO stages</span></h2> <p data-svelte-h="svelte-7cwums">Select a ZeRO stage config to use as a starting point.</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">ZeRO-1 </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">ZeRO-2 </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">ZeRO-3 </div></div> <div class="language-select"><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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;bf16&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;enabled&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span> <span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;zero_optimization&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;stage&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">1</span> <span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;gradient_clipping&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;train_micro_batch_size_per_gpu&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;train_batch_size&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;gradient_accumulation_steps&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span>
<span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div> </div> <p data-svelte-h="svelte-j937oq">The following fields are important for customizing training.</p> <ul><li><p data-svelte-h="svelte-11r066v"><code>zero_optimization</code> sets the ZeRO stage.</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;zero_optimization&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;stage&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">3</span> <span class="hljs-punctuation">}</span> <span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div></li> <li><p data-svelte-h="svelte-1jqjzaf">Set the batch size and gradient accumulation arguments to <code>&quot;auto&quot;</code>. If you manually set these to values that disagree with <a href="/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a>, training continues silently with the wrong values.</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;train_micro_batch_size_per_gpu&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;train_batch_size&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;gradient_accumulation_steps&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;gradient_clipping&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span>
<span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div></li> <li><p data-svelte-h="svelte-1ysp1gv"><code>bf16</code> sets the training precision. Set it to <code>&quot;auto&quot;</code> so it mirrors the <code>bf16</code> flag in <a href="/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;bf16&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;enabled&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span> <span class="hljs-punctuation">}</span> <span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div></li> <li><p data-svelte-h="svelte-1y9s3aq"><code>stage3_gather_16bit_weights_on_model_save</code> performs an all-gather across all GPUs before saving, reconstructing the full tensors from their shards. This is a ZeRO-3 argument.</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;zero_optimization&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;stage&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">3</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;stage3_gather_16bit_weights_on_model_save&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-literal"><span class="hljs-keyword">true</span></span><span class="hljs-punctuation">,</span>
 <span class="hljs-punctuation">}</span>
<span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div></li> <li><p data-svelte-h="svelte-is3i0d">Set <code>overlap_comm</code> to <code>true</code> to hide all-reduce latency behind the backward pass. <code>allgather_bucket_size</code> and <code>reduce_bucket_size</code> trade communication speed for GPU memory. Lower values use less memory but slow communication.</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;zero_optimization&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;stage&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">2</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;overlap_comm&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-literal"><span class="hljs-keyword">true</span></span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;allgather_bucket_size&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">2e8</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;reduce_bucket_size&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">2e8</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;contiguous_gradients&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-literal"><span class="hljs-keyword">true</span></span>
 <span class="hljs-punctuation">}</span>
<span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div></li> <li><p data-svelte-h="svelte-6md6qc"><code>offload_optimizer</code> offloads the optimizer to CPU memory. To save even more memory, also offload model parameters with <code>offload_param</code> (ZeRO-3 only). Set <code>pin_memory</code> to <code>true</code> to speed up CPU-GPU transfers, but this locks RAM that is unavailable to other processes.</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;zero_optimization&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;stage&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-number">3</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;offload_optimizer&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;device&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;cpu&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;pin_memory&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-literal"><span class="hljs-keyword">true</span></span> <span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;offload_param&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;device&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;cpu&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;pin_memory&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-literal"><span class="hljs-keyword">true</span></span> <span class="hljs-punctuation">}</span>
 <span class="hljs-punctuation">}</span>
<span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div></li> <li><p data-svelte-h="svelte-vqyguf"><code>optimizer</code> and <code>scheduler</code> default to the optimizer and scheduler configured in <a href="/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a>. Set to <code>&quot;auto&quot;</code> so DeepSpeed reads the values from <a href="/docs/transformers/main/en/main_classes/trainer#transformers.TrainingArguments">TrainingArguments</a> unless you need a DeepSpeed-native optimizer like LAMB.</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;optimizer&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;type&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;AdamW&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;params&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;lr&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;betas&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;eps&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;weight_decay&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span> <span class="hljs-punctuation">}</span>
 <span class="hljs-punctuation">}</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;scheduler&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;type&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;WarmupDecayLR&quot;</span><span class="hljs-punctuation">,</span>
 <span class="hljs-attr">&quot;params&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-punctuation">{</span> <span class="hljs-attr">&quot;total_num_steps&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;warmup_min_lr&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;warmup_max_lr&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span><span class="hljs-punctuation">,</span> <span class="hljs-attr">&quot;warmup_num_steps&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-string">&quot;auto&quot;</span> <span class="hljs-punctuation">}</span>
 <span class="hljs-punctuation">}</span>
<span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-lazfz1">If you’re offloading the optimizer, set <code>zero_force_ds_cpu_optimizer</code> to <code>false</code> to use DeepSpeed’s CPU Adam optimizer.</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="language-json "><!-- HTML_TAG_START --><span class="hljs-punctuation">{</span>
 <span class="hljs-attr">&quot;zero_force_ds_cpu_optimizer&quot;</span><span class="hljs-punctuation">:</span> <span class="hljs-literal"><span class="hljs-keyword">false</span></span>
<span class="hljs-punctuation">}</span><!-- HTML_TAG_END --></pre></div></li></ul> <h2 class="relative group"><a id="checkpoints" 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="#checkpoints"><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>Checkpoints</span></h2> <p data-svelte-h="svelte-4yz1kz">DeepSpeed saves checkpoints in a sharded format that can’t be loaded directly with <a href="/docs/transformers/main/en/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a>. Set <code>load_best_model_at_end()</code> to <code>True</code> to have Trainer track and reload the best checkpoint at the end of training.</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="language-py "><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> TrainingArguments, Trainer
args = TrainingArguments(
 deepspeed=<span class="hljs-string">&quot;ds_config_zero3.json&quot;</span>,
 load_best_model_at_end=<span class="hljs-literal">True</span>,
 ...
)
<span class="hljs-comment"># after training, save a normal transformers checkpoint</span>
trainer.save_model(<span class="hljs-string">&quot;./best-model&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-zifc4j">Setting <code>save_only_model=True</code> skips saving the full optimizer state, which means you can’t reload the best model at the end of training. Also set <code>stage3_gather_16bit_weights_on_model_save: true</code> to reconstruct full weights from their shards. This is required for saving a consolidated 16-bit model artifact or 16-bit state dict with ZeRO-3. Transformers raises an error when <code>save_only_model=True</code> is combined with <code>load_best_model_at_end=True</code>.</p> <blockquote class="tip" data-svelte-h="svelte-12ekg4t"><p>For resuming across different parallelism configurations, see DeepSpeed’s <a href="https://www.deepspeed.ai/tutorials/universal-checkpointing" rel="nofollow">Universal Checkpointing</a> guide.</p></blockquote> <h2 class="relative group"><a id="next-steps" 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="#next-steps"><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>Next steps</span></h2> <ul data-svelte-h="svelte-1kmes8g"><li>Read the <a href="https://nanotron-ultrascale-playbook.static.hf.space/index.html#zero_redundancy_optimizer_(zero)" rel="nofollow">Zero Redundancy Optimizer</a> chapter from The Ultra-Scale Playbook to learn more about how ZeRO works.</li> <li>Read the ZeRO papers: <a href="https://hf.co/papers/1910.02054" rel="nofollow">Memory Optimizations Toward Training Trillion Parameter Models</a>, <a href="https://hf.co/papers/2101.06840" rel="nofollow">Democratizing Billion-Scale Model Training</a>, and <a href="https://hf.co/papers/2104.07857" rel="nofollow">Breaking the GPU Memory Wall for Extreme Scale Deep Learning</a>.</li></ul> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/transformers/blob/main/docs/source/en/deepspeed.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_rv114u = {
 assets: "/docs/transformers/main/en",
 base: "/docs/transformers/main/en",
 env: {}
 };
 const element = document.currentScript.parentElement;
 const data = [null,null];
 Promise.all([
 import("/docs/transformers/main/en/_app/immutable/entry/start.3d6cca8a.js"),
 import("/docs/transformers/main/en/_app/immutable/entry/app.b6ccab0d.js")
 ]).then(([kit, app]) => {
 kit.start(app, element, {
 node_ids: [0, 43],
 data,
 form: null,
 error: null
 });
 });
 }
 </script>

Xet Storage Details

Size:
53.9 kB
·
Xet hash:
44f51e90e8482f73d6b238f839a41c6aa43cd3d7e51890e4700510cf6be3a6df

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