Hugging Face's logo

Buckets:
hf-doc-build
/
doc-dev

Files
xet
hf-doc-build/doc-dev / accelerate /main /en /usage_guides /ddp_comm_hook.html
rtrm's picture
rtrm
download
raw
30.9 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;DDP Communication Hooks&quot;,&quot;local&quot;:&quot;ddp-communication-hooks&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;FP16 Compression Hook&quot;,&quot;local&quot;:&quot;fp16-compression-hook&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;BF16 Compression Hook&quot;,&quot;local&quot;:&quot;bf16-compression-hook&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;PowerSGD Hook&quot;,&quot;local&quot;:&quot;powersgd-hook&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;DDP Communication Hooks utilities&quot;,&quot;local&quot;:&quot;ddp-communication-hooks-utilities&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;comm_wrapper&quot;,&quot;local&quot;:&quot;commwrapper&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;comm_state_option&quot;,&quot;local&quot;:&quot;commstateoption&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
 <link href="/docs/accelerate/main/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/entry/start.2ea03080.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/scheduler.defa9a21.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/singletons.aff0b9fc.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/index.beade68d.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/paths.2c85d1a6.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/entry/app.e6812672.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/index.fe795e71.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/nodes/0.39c84d5d.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/each.e59479a4.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/nodes/37.3c85add2.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/Tip.179eb360.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/CodeBlock.42404125.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/EditOnGithub.0f575778.js">
 <link rel="modulepreload" href="/docs/accelerate/main/en/_app/immutable/chunks/HfOption.3c290b0f.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;DDP Communication Hooks&quot;,&quot;local&quot;:&quot;ddp-communication-hooks&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;FP16 Compression Hook&quot;,&quot;local&quot;:&quot;fp16-compression-hook&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;BF16 Compression Hook&quot;,&quot;local&quot;:&quot;bf16-compression-hook&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;PowerSGD Hook&quot;,&quot;local&quot;:&quot;powersgd-hook&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2},{&quot;title&quot;:&quot;DDP Communication Hooks utilities&quot;,&quot;local&quot;:&quot;ddp-communication-hooks-utilities&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;comm_wrapper&quot;,&quot;local&quot;:&quot;commwrapper&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3},{&quot;title&quot;:&quot;comm_state_option&quot;,&quot;local&quot;:&quot;commstateoption&quot;,&quot;sections&quot;:[],&quot;depth&quot;:3}],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="ddp-communication-hooks" 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="#ddp-communication-hooks"><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>DDP Communication Hooks</span></h1> <p data-svelte-h="svelte-x6tyko">Distributed Data Parallel (DDP) communication hooks provide a generic interface to control how gradients are communicated across workers by overriding the vanilla allreduce in <code>DistributedDataParallel</code>. A few built-in communication hooks are provided, and users can easily apply any of these hooks to optimize communication.</p> <ul data-svelte-h="svelte-zxp3oz"><li><strong>FP16 Compression Hook</strong>: Compresses gradients by casting them to half-precision floating-point format (<code>torch.float16</code>), reducing communication overhead.</li> <li><strong>BF16 Compression Hook</strong>: Similar to FP16, but uses the Brain Floating Point format (<code>torch.bfloat16</code>), which can be more efficient on certain hardware.</li> <li><strong>PowerSGD Hook</strong>: An advanced gradient compression algorithm that provides high compression rates and can accelerate bandwidth-bound distributed training.</li></ul> <p data-svelte-h="svelte-13wlhyn">In this tutorial, you will see how to quickly set up DDP communication hooks and perform training with the utilities provided in 🤗 Accelerate, which can be as simple as adding just one new line of code! This demonstrates how to use DDP communication hooks to optimize gradient communication in distributed training with the 🤗 Accelerate library.</p> <h2 class="relative group"><a id="fp16-compression-hook" 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="#fp16-compression-hook"><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>FP16 Compression Hook</span></h2> <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">PyTorch </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 </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=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> torch.nn.parallel <span class="hljs-keyword">import</span> DistributedDataParallel <span class="hljs-keyword">as</span> DDP
<span class="hljs-keyword">from</span> torch.distributed.algorithms.ddp_comm_hooks <span class="hljs-keyword">import</span> default_hooks
<span class="hljs-keyword">class</span> <span class="hljs-title class_">MyModel</span>(torch.nn.Module):
 <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self</span>):
 <span class="hljs-built_in">super</span>().__init__()
 self.layer = torch.nn.Linear(<span class="hljs-number">10</span>, <span class="hljs-number">10</span>)
<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, x</span>):
 <span class="hljs-keyword">return</span> self.layer(x)
model = MyModel()
model = DDP(model, device_ids=[torch.cuda.current_device()])
model.register_comm_hook(state=<span class="hljs-literal">None</span>, hook=default_hooks.fp16_compress_hook)
<span class="hljs-comment"># Training loop</span>
<span class="hljs-keyword">for</span> data, targets <span class="hljs-keyword">in</span> data_loader:
 outputs = model(data)
 loss = criterion(outputs, targets)
 loss.backward()
 optimizer.step()
 optimizer.zero_grad()<!-- HTML_TAG_END --></pre></div> </div> <h3 class="relative group"><a id="bf16-compression-hook" 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="#bf16-compression-hook"><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>BF16 Compression Hook</span></h3> <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-1s0p8uc">BF16 Compression Hook API is experimental, and it requires NCCL version later than 2.9.6.</p></div> <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">PyTorch </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 </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=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> torch.nn.parallel <span class="hljs-keyword">import</span> DistributedDataParallel <span class="hljs-keyword">as</span> DDP
<span class="hljs-keyword">from</span> torch.distributed.algorithms.ddp_comm_hooks <span class="hljs-keyword">import</span> default_hooks
<span class="hljs-keyword">class</span> <span class="hljs-title class_">MyModel</span>(torch.nn.Module):
 <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self</span>):
 <span class="hljs-built_in">super</span>().__init__()
 self.layer = torch.nn.Linear(<span class="hljs-number">10</span>, <span class="hljs-number">10</span>)
<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, x</span>):
 <span class="hljs-keyword">return</span> self.layer(x)
model = MyModel()
model = DDP(model, device_ids=[torch.cuda.current_device()])
model.register_comm_hook(state=<span class="hljs-literal">None</span>, hook=default_hooks.bf16_compress_hook)
<span class="hljs-comment"># Training loop</span>
<span class="hljs-keyword">for</span> data, targets <span class="hljs-keyword">in</span> data_loader:
 outputs = model(data)
 loss = criterion(outputs, targets)
 loss.backward()
 optimizer.step()
 optimizer.zero_grad()<!-- HTML_TAG_END --></pre></div> </div> <h3 class="relative group"><a id="powersgd-hook" 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="#powersgd-hook"><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>PowerSGD Hook</span></h3> <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-1f2etf0">PowerSGD typically requires extra memory of the same size as the model’s gradients to enable error feedback, which can compensate for biased compressed communication and improve accuracy.</p></div> <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">PyTorch </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 </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=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> torch.nn.parallel <span class="hljs-keyword">import</span> DistributedDataParallel <span class="hljs-keyword">as</span> DDP
<span class="hljs-keyword">from</span> torch.distributed.algorithms.ddp_comm_hooks <span class="hljs-keyword">import</span> powerSGD_hook
<span class="hljs-keyword">class</span> <span class="hljs-title class_">MyModel</span>(torch.nn.Module):
 <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self</span>):
 <span class="hljs-built_in">super</span>().__init__()
 self.layer = torch.nn.Linear(<span class="hljs-number">10</span>, <span class="hljs-number">10</span>)
<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, x</span>):
 <span class="hljs-keyword">return</span> self.layer(x)
model = MyModel()
model = DDP(model, device_ids=[torch.cuda.current_device()])
state = powerSGD_hook.PowerSGDState(process_group=<span class="hljs-literal">None</span>)
model.register_comm_hook(state=state, hook=powerSGD_hook.powerSGD_hook)
<span class="hljs-comment"># Training loop</span>
<span class="hljs-keyword">for</span> data, targets <span class="hljs-keyword">in</span> data_loader:
 outputs = model(data)
 loss = criterion(outputs, targets)
 loss.backward()
 optimizer.step()
 optimizer.zero_grad()<!-- HTML_TAG_END --></pre></div> </div> <h2 class="relative group"><a id="ddp-communication-hooks-utilities" 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="#ddp-communication-hooks-utilities"><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>DDP Communication Hooks utilities</span></h2> <p data-svelte-h="svelte-1nbipsx">There are two additional utilities for supporting optional functionalities with the communication hooks.</p> <h3 class="relative group"><a id="commwrapper" 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="#commwrapper"><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>comm_wrapper</span></h3> <p data-svelte-h="svelte-oi3r9p"><code>comm_wrapper</code> is an option to wrap a communication hook with additional functionality. For example, it can be used to combine FP16 compression with other communication strategies. Currently supported wrappers are <code>no</code>, <code>fp16</code>, and <code>bf16</code>.</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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> accelerate <span class="hljs-keyword">import</span> Accelerator, DDPCommunicationHookType, DistributedDataParallelKwargs
<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">class</span> <span class="hljs-title class_">MyModel</span>(torch.nn.Module):
 <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self</span>):
 <span class="hljs-built_in">super</span>().__init__()
 self.layer = torch.nn.Linear(<span class="hljs-number">10</span>, <span class="hljs-number">10</span>)
<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, x</span>):
 <span class="hljs-keyword">return</span> self.layer(x)
<span class="hljs-comment"># DDP Communication Hook setup</span>
ddp_kwargs = DistributedDataParallelKwargs(
 comm_hook=DDPCommunicationHookType.POWER_SGD,
 comm_wrapper=DDPCommunicationHookType.FP16
)
accelerator = Accelerator(kwargs_handlers=[ddp_kwargs])
model = MyModel()
optimizer = torch.optim.Adam(model.parameters())
data_loader = DataLoader(dataset, batch_size=<span class="hljs-number">16</span>)
model, optimizer, data_loader = accelerator.prepare(model, optimizer, data_loader)
<span class="hljs-comment"># Training loop</span>
<span class="hljs-keyword">for</span> data, targets <span class="hljs-keyword">in</span> data_loader:
 outputs = model(data)
 loss = criterion(outputs, targets)
 accelerator.backward(loss)
 optimizer.step()
 optimizer.zero_grad()<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="commstateoption" 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="#commstateoption"><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>comm_state_option</span></h3> <p data-svelte-h="svelte-1igm2b3"><code>comm_state_option</code> allows you to pass additional state information required by certain communication hooks. This is particularly useful for stateful hooks like <code>PowerSGD</code>, which require maintaining hyperparameters and internal states across training steps. Below is an example showcasing the use of <code>comm_state_option</code> with the <code>PowerSGD</code> hook.</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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> accelerate <span class="hljs-keyword">import</span> Accelerator, DDPCommunicationHookType, DistributedDataParallelKwargs
<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">class</span> <span class="hljs-title class_">MyModel</span>(torch.nn.Module):
 <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self</span>):
 <span class="hljs-built_in">super</span>().__init__()
 self.layer = torch.nn.Linear(<span class="hljs-number">10</span>, <span class="hljs-number">10</span>)
<span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, x</span>):
 <span class="hljs-keyword">return</span> self.layer(x)
<span class="hljs-comment"># DDP Communication Hook setup</span>
ddp_kwargs = DistributedDataParallelKwargs(
 comm_hook=DDPCommunicationHookType.POWER_SGD,
 comm_state_option={<span class="hljs-string">&quot;matrix_approximation_rank&quot;</span>: <span class="hljs-number">2</span>}
)
accelerator = Accelerator(kwargs_handlers=[ddp_kwargs])
model = MyModel()
optimizer = torch.optim.Adam(model.parameters())
data_loader = DataLoader(dataset, batch_size=<span class="hljs-number">16</span>)
model, optimizer, data_loader = accelerator.prepare(model, optimizer, data_loader)
<span class="hljs-comment"># Training loop</span>
<span class="hljs-keyword">for</span> data, targets <span class="hljs-keyword">in</span> data_loader:
 outputs = model(data)
 loss = criterion(outputs, targets)
 accelerator.backward(loss)
 optimizer.step()
 optimizer.zero_grad()<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1ll6vn">For more advanced usage and additional hooks, refer to the <a href="https://pytorch.org/docs/stable/ddp_comm_hooks.html" rel="nofollow">PyTorch DDP Communication Hooks documentation</a>.</p> <a class="!text-gray-400 !no-underline text-sm flex items-center not-prose mt-4" href="https://github.com/huggingface/accelerate/blob/main/docs/source/usage_guides/ddp_comm_hook.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_1fyccrg = {
 assets: "/docs/accelerate/main/en",
 base: "/docs/accelerate/main/en",
 env: {}
 };
 const element = document.currentScript.parentElement;
 const data = [null,null];
 Promise.all([
 import("/docs/accelerate/main/en/_app/immutable/entry/start.2ea03080.js"),
 import("/docs/accelerate/main/en/_app/immutable/entry/app.e6812672.js")
 ]).then(([kit, app]) => {
 kit.start(app, element, {
 node_ids: [0, 37],
 data,
 form: null,
 error: null
 });
 });
 }
 </script>

Xet Storage Details

Size:
30.9 kB
·
Xet hash:
54c0eeee7d57d5551929323596d9253329e8ec75f2f7db35f0e2e0586c41721c

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