Buckets:

hf-doc-build
/

doc

Files
xet
hf-doc-build/doc / transformers /main /pt /kernel_doc /writing_kernels.html
mishig
3 days ago
download
raw
31.6 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Escrevendo kernels&quot;,&quot;local&quot;:&quot;writing-kernels&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Padrão de duas classes&quot;,&quot;local&quot;:&quot;two-class-pattern&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Transformação de parâmetros&quot;,&quot;local&quot;:&quot;parameter-transformation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Fusão de módulos&quot;,&quot;local&quot;:&quot;module-fusion&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/>
		<link href="/docs/transformers/main/en/_app/immutable/entry/start.ByZEd4qm.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/DHdg-_0Z.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/CBySwR1V.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/entry/app.CQqZS61p.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/Cdb1BouH.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/dUGfULTE.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/DOLBYqcr.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/DsnmJJEf.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/nodes/0.CEy2b_T8.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/Conj1S5Z.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/nodes/88.Br6wh_o6.js" rel="modulepreload">
		<link href="/docs/transformers/main/en/_app/immutable/chunks/DoQ7MZQh.js" rel="modulepreload">
		<!--dlfaz7--><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Writing kernels&quot;,&quot;local&quot;:&quot;writing-kernels&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;Two-class pattern&quot;,&quot;local&quot;:&quot;two-class-pattern&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Parameter transformation&quot;,&quot;local&quot;:&quot;parameter-transformation&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;Module fusion&quot;,&quot;local&quot;:&quot;module-fusion&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"/><!---->
		<link href="/docs/transformers/main/en/_app/immutable/assets/0.tn0RQdqM.css" rel="modulepreload"> <!--[--><!--[0--><!--[--><!--[0--><!--[--><p></p> <div class="items-center shrink-0 min-w-[100px] max-sm:min-w-[50px] justify-end ml-auto flex" style="float: inline-end; margin-inline-start: 10px; display: inline-flex; position: relative; z-index: 10;"><div dir="ltr" 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 dir="auto">Copy page</span></button> <button class="inline-flex items-center justify-center w-6 max-sm:w-5 h-7 max-sm:h-7 disabled:pointer-events-none text-sm text-gray-500 hover:text-gray-700 dark:hover:text-white rounded-r-md max-sm:rounded-r-sm border border-l transition border-gray-200 bg-white hover:shadow-inner dark:border-gray-850 dark:bg-gray-950 dark:text-gray-200 dark:hover:bg-gray-800" aria-haspopup="menu" aria-expanded="false" aria-label="Open copy menu"><svg class="transition-transform text-gray-400 overflow-visible sm:size-3.5 size-3 rotate-0" width="1em" height="1em" viewBox="0 0 12 7" fill="none" xmlns="http://www.w3.org/2000/svg"><path d="M1 1L6 6L11 1" stroke="currentColor"></path></svg><!----></button></div> <!--[-1--><!--]--></div><!----> <!--[0--><h1 class="relative group"><a id="writing-kernels" 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="#writing-kernels"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Escrevendo kernels<!----></span></h1><!--]--><!----> <p>Este guia explica como escrever kernels que vão além de uma substituição <code>forward</code> sem estado. Ele aborda duas capacidades suportadas pela API estendida <code>KernelConfig</code>:</p> <ol><li>Transformação de parâmetros: o kernel espera pesos em um layout diferente do modelo original (por exemplo, parâmetros renomeados ou mesclados).</li> <li>Fusão de módulos: o kernel substitui múltiplos módulos adjacentes por uma única implementação fundida.</li></ol> <p>Para kernels básicos (substituições de <code>forward</code> sem estado, sem alterações de parâmetros), consulte a documentação da biblioteca <a href="https://github.com/huggingface/kernels" rel="nofollow">kernels</a>.</p> <!--[1--><h2 class="relative group"><a id="two-class-pattern" 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="#two-class-pattern"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Padrão de duas classes<!----></span></h2><!--]--><!----> <p>Qualquer kernel que carrega seus próprios parâmetros segue um padrão de duas classes.</p> <ul><li><code>KernelName</code>: contém apenas a passagem <code>forward</code>. A biblioteca <code>kernels</code> usa essa classe para kernelizar o modelo, pois não permite classes de kernel com estado.</li> <li><code>KernelNameLayout</code>: um <code>nn.Module</code> que armazena os parâmetros e aplica monkey-patching no módulo original antes que o checkpoint seja carregado. Em tempo de execução, <code>kernelize</code> substitui seu <code>forward</code> pelo <code>forward</code> de <code>KernelName</code>. Você não precisa definir <code>forward</code>. O Transformers injeta um automaticamente com a mesma assinatura de <code>KernelName.forward</code>.</li></ul> <blockquote class="important"><p>A convenção de nomenclatura é rigorosa. A classe de layout deve ser nomeada <code>{KernelName}Layout</code> e definida no mesmo módulo de <code>KernelName</code>.</p></blockquote> <!--[1--><h2 class="relative group"><a id="parameter-transformation" 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="#parameter-transformation"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Transformação de parâmetros<!----></span></h2><!--]--><!----> <p>Use este padrão quando o kernel espera pesos com nomes diferentes ou em uma forma diferente da do checkpoint original do modelo.</p> <p>A classe <code>KernelNameLayout</code> possui a mesma assinatura de <code>__init__</code> que o módulo que substitui e declara um atributo de classe <code>conversion_mapping</code> que informa ao Transformers como remapear as chaves do checkpoint para os novos nomes de parâmetros (veja <a href="../weightconverter">Carregamento dinâmico de pesos</a> para mais detalhes).</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 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> Copiado</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-comment"># layers.py</span>
<span class="hljs-keyword">try</span>:
    <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Concatenate, WeightConverter
    <span class="hljs-keyword">from</span> transformers.conversion_mapping <span class="hljs-keyword">import</span> WeightRenaming
<span class="hljs-keyword">except</span> ImportError:
    Concatenate = WeightConverter = WeightRenaming = <span class="hljs-literal">None</span>

<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">import</span> torch.nn <span class="hljs-keyword">as</span> nn


<span class="hljs-keyword">class</span> <span class="hljs-title class_">CustomRMSNormLayout</span>(nn.Module):
    conversion_mapping = [...]  <span class="hljs-comment"># rules that remap checkpoint keys to the new parameter names</span>

    <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self, hidden_size: <span class="hljs-built_in">int</span>, eps: <span class="hljs-built_in">float</span> = <span class="hljs-number">1e-6</span></span>):
        <span class="hljs-built_in">super</span>().__init__()
        <span class="hljs-variable language_">self</span>.scale = nn.Parameter(torch.ones(hidden_size))
        <span class="hljs-variable language_">self</span>.variance_epsilon = eps


<span class="hljs-keyword">class</span> <span class="hljs-title class_">CustomRMSNorm</span>(nn.Module):
    <span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, hidden_states: torch.Tensor</span>) -&gt; torch.Tensor:
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.<span class="hljs-built_in">pow</span>(<span class="hljs-number">2</span>).mean(-<span class="hljs-number">1</span>, keepdim=<span class="hljs-literal">True</span>)
        hidden_states = hidden_states * torch.rsqrt(variance + <span class="hljs-variable language_">self</span>.variance_epsilon)
        <span class="hljs-keyword">return</span> <span class="hljs-variable language_">self</span>.scale * hidden_states.to(input_dtype)<!----></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 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> Copiado</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-comment"># __init__.py</span>
<span class="hljs-keyword">from</span> . <span class="hljs-keyword">import</span> layers<!----></pre></div><!----> <blockquote class="note"><p>A biblioteca <code>kernels</code> descobre as classes de kernel por meio de um módulo <code>layers</code> (seja <code>layers.py</code> ou <code>layers/__init__.py</code>). Defina as classes de kernel e de layout diretamente em <code>layers.py</code> e exponha o módulo <code>layers</code> a partir do <code>__init__.py</code> do pacote. O Transformers encontra <code>CustomRMSNormLayout</code> no mesmo módulo de <code>CustomRMSNorm</code>.</p> <p>Proteja as importações específicas do Transformers (<code>WeightConverter</code>, <code>Concatenate</code>, <code>WeightRenaming</code>) em <code>layers.py</code> para que o kernel também possa ser importado em ambientes em que o Transformers não está instalado.</p></blockquote> <p>Carregue este kernel passando o repositório e o nome da classe para <a href="/docs/transformers/main/pt/main_classes/kernels#transformers.KernelConfig">KernelConfig</a>. A chave é o nome da classe do módulo original do modelo. O valor aponta para a classe <code>KernelName</code> (não para o <code>Layout</code>) no repositório.</p> <p>O valor é uma string <code>"owner/repo:ClassName"</code> ou uma tupla <code>(repo_string, metadata)</code>. No formato de tupla, passe <code>version</code> ou <code>revision</code> (pelo menos um) e, opcionalmente, <code>trust_remote_code</code>. Use <code>version</code> para uma versão principal do kernel ou <code>revision</code> para uma tag, branch ou commit. Defina <code>trust_remote_code=True</code> para carregar de fora da organização confiável <code>kernels-community</code>. Uma string simples tem como padrão <code>version=1</code> e <code>trust_remode_code=False</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 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> Copiado</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, KernelConfig

kernel_config = KernelConfig({<span class="hljs-string">&quot;RMSNorm&quot;</span>: (<span class="hljs-string">&quot;owner/my-kernel:CustomRMSNorm&quot;</span>, {<span class="hljs-string">&quot;trust_remote_code&quot;</span>: <span class="hljs-literal">True</span>, <span class="hljs-string">&quot;version&quot;</span>: <span class="hljs-number">1</span>})})
model = AutoModelForCausalLM.from_pretrained(
    <span class="hljs-string">&quot;Qwen/Qwen3-0.6B&quot;</span>,
    use_kernels=<span class="hljs-literal">True</span>,
    kernel_config=kernel_config,
    device_map=<span class="hljs-string">&quot;auto&quot;</span>,
)<!----></pre></div><!----> <p>Quando o modelo é carregado, o Transformers:</p> <ol><li>Carrega <code>CustomRMSNorm</code> do repositório e procura por <code>CustomRMSNormLayout</code> no mesmo módulo.</li> <li>Aplica monkey-patch em cada <code>RMSNorm</code> do modelo com <code>CustomRMSNormLayout</code>.</li> <li>Remapeia os pesos do checkpoint usando <code>conversion_mapping</code> para que sejam carregados nos novos nomes de parâmetros.</li> <li>Chama <code>kernelize</code>, que substitui <code>CustomRMSNormLayout.forward</code> por <code>CustomRMSNorm.forward</code>.</li></ol> <!--[1--><h2 class="relative group"><a id="module-fusion" 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="#module-fusion"><span><svg xmlns="http://www.w3.org/2000/svg" xmlns:xlink="http://www.w3.org/1999/xlink" aria-hidden="true" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 256 256"><path d="M167.594 88.393a8.001 8.001 0 0 1 0 11.314l-67.882 67.882a8 8 0 1 1-11.314-11.315l67.882-67.881a8.003 8.003 0 0 1 11.314 0zm-28.287 84.86l-28.284 28.284a40 40 0 0 1-56.567-56.567l28.284-28.284a8 8 0 0 0-11.315-11.315l-28.284 28.284a56 56 0 0 0 79.196 79.197l28.285-28.285a8 8 0 1 0-11.315-11.314zM212.852 43.14a56.002 56.002 0 0 0-79.196 0l-28.284 28.284a8 8 0 1 0 11.314 11.314l28.284-28.284a40 40 0 0 1 56.568 56.567l-28.285 28.285a8 8 0 0 0 11.315 11.314l28.284-28.284a56.065 56.065 0 0 0 0-79.196z" fill="currentColor"></path></svg><!----></span></a> <span><!---->Fusão de módulos<!----></span></h2><!--]--><!----> <p>Use este padrão quando um kernel substitui vários módulos adjacentes por uma única implementação fundida. Como o módulo fundido combina parâmetros de vários módulos originais, o <code>KernelNameLayout.__init__</code> recebe os módulos filhos instanciados, em vez dos argumentos de seus construtores.</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 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> Copiado</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-comment"># layers.py</span>
<span class="hljs-keyword">try</span>:
    <span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> Concatenate, WeightConverter
    <span class="hljs-keyword">from</span> transformers.conversion_mapping <span class="hljs-keyword">import</span> WeightRenaming
<span class="hljs-keyword">except</span> ImportError:
    Concatenate = WeightConverter = WeightRenaming = <span class="hljs-literal">None</span>

<span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">import</span> torch.nn <span class="hljs-keyword">as</span> nn


<span class="hljs-keyword">class</span> <span class="hljs-title class_">RMSNormMLPLayout</span>(nn.Module):
    conversion_mapping = [...]  <span class="hljs-comment"># rules that remap checkpoint keys to the fused parameter names</span>

    <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self, norm, mlp</span>):
        <span class="hljs-built_in">super</span>().__init__()
        <span class="hljs-variable language_">self</span>.variance_epsilon = norm.variance_epsilon
        <span class="hljs-variable language_">self</span>.scale = nn.Parameter(torch.empty_like(norm.weight))
        <span class="hljs-variable language_">self</span>.gate_up_proj = nn.Linear(
            mlp.gate_proj.in_features,
            mlp.gate_proj.out_features + mlp.up_proj.out_features,
            bias=mlp.gate_proj.bias <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">None</span>,
            device=mlp.gate_proj.weight.device,
            dtype=mlp.gate_proj.weight.dtype,
        )
        <span class="hljs-variable language_">self</span>.down_proj = nn.Linear(
            mlp.down_proj.in_features,
            mlp.down_proj.out_features,
            bias=mlp.down_proj.bias <span class="hljs-keyword">is</span> <span class="hljs-keyword">not</span> <span class="hljs-literal">None</span>,
            device=mlp.down_proj.weight.device,
            dtype=mlp.down_proj.weight.dtype,
        )
        <span class="hljs-variable language_">self</span>.act_fn = mlp.act_fn


<span class="hljs-keyword">class</span> <span class="hljs-title class_">RMSNormMLP</span>(nn.Module):
    <span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, hidden_states: torch.Tensor</span>) -&gt; torch.Tensor:
        input_dtype = hidden_states.dtype
        hidden_states = hidden_states.to(torch.float32)
        variance = hidden_states.<span class="hljs-built_in">pow</span>(<span class="hljs-number">2</span>).mean(-<span class="hljs-number">1</span>, keepdim=<span class="hljs-literal">True</span>)
        hidden_states = hidden_states * torch.rsqrt(variance + <span class="hljs-variable language_">self</span>.variance_epsilon)
        hidden_states = <span class="hljs-variable language_">self</span>.scale * hidden_states.to(input_dtype)
        gate, up = <span class="hljs-variable language_">self</span>.gate_up_proj(hidden_states).chunk(<span class="hljs-number">2</span>, dim=-<span class="hljs-number">1</span>)
        <span class="hljs-keyword">return</span> <span class="hljs-variable language_">self</span>.down_proj(<span class="hljs-variable language_">self</span>.act_fn(gate) * up)<!----></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 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> Copiado</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-comment"># __init__.py</span>
<span class="hljs-keyword">from</span> . <span class="hljs-keyword">import</span> layers<!----></pre></div><!----> <p>Para fundir módulos, passe uma tupla de pares <code>(class_name, path_pattern)</code> como chave em <code>KernelConfig</code> em vez de uma string simples. Todos os padrões devem compartilhar o mesmo módulo pai (o Transformers funde os filhos nesse pai). O curinga <code>*</code> corresponde a qualquer segmento único de caminho.</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 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> Copiado</div><!----></button><!----></div> <pre class="language-python "><!----><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, KernelConfig

kernel_config = KernelConfig(
    {
        (
            (<span class="hljs-string">&quot;RMSNorm&quot;</span>, <span class="hljs-string">&quot;model.layers.*.post_attention_layernorm&quot;</span>),
            (<span class="hljs-string">&quot;MLP&quot;</span>,     <span class="hljs-string">&quot;model.layers.*.mlp&quot;</span>),
        ): (<span class="hljs-string">&quot;owner/my-kernel:RMSNormMLP&quot;</span>, {<span class="hljs-string">&quot;trust_remote_code&quot;</span>: <span class="hljs-literal">True</span>, <span class="hljs-string">&quot;version&quot;</span>: <span class="hljs-number">1</span>}),
    }
)
model = AutoModelForCausalLM.from_pretrained(
    <span class="hljs-string">&quot;Qwen/Qwen3-0.6B&quot;</span>,
    use_kernels=<span class="hljs-literal">True</span>,
    kernel_config=kernel_config,
    device_map=<span class="hljs-string">&quot;auto&quot;</span>,
)<!----></pre></div><!----> <p>Quando o modelo é carregado, o Transformers:</p> <ol><li>Carrega <code>RMSNormMLP</code> do repositório e localiza <code>RMSNormMLPLayout</code> no mesmo módulo.</li> <li>Corresponde a cada camada do decoder em <code>model.layers.*</code> e constrói uma classe pai fundida cujo <code>__init__</code> chama <code>RMSNormMLPLayout(post_attention_layernorm, mlp)</code>.</li> <li>Substitui o filho restante (<code>mlp</code>) por <code>nn.Identity()</code> para preservar a interface do módulo pai.</li> <li>Remapeia os pesos do checkpoint usando <code>conversion_mapping</code>.</li> <li>Chama <code>kernelize</code>, que substitui <code>RMSNormMLPLayout.forward</code> por <code>RMSNormMLP.forward</code>.</li></ol> <blockquote class="tip"><p>A ordem dos pares na tupla de fusão determina a ordem dos argumentos passados para <code>KernelNameLayout.__init__</code>.</p></blockquote> <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/kernel_doc/writing_kernels.md" target="_blank"><svg class="mr-1" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M31,16l-7,7l-1.41-1.41L28.17,16l-5.58-5.59L24,9l7,7z"></path><path d="M1,16l7-7l1.41,1.41L3.83,16l5.58,5.59L8,23l-7-7z"></path><path d="M12.419,25.484L17.639,6.552l1.932,0.518L14.351,26.002z"></path></svg><!----> <span><span class="underline">Update</span> on GitHub</span></a><!----> <p></p><!--]--><!----><!--]--><!--]--><!--]--> <!--[-1--><!--]--><!--]-->
			
			<script>
				{
					__sveltekit_1y3v6y9 = {
						base: "/docs/transformers/main/en",
						assets: "/docs/transformers/main/en"
					};
					const element = document.currentScript.parentElement;
					Promise.all([
						import("/docs/transformers/main/en/_app/immutable/entry/start.ByZEd4qm.js"),
						import("/docs/transformers/main/en/_app/immutable/entry/app.CQqZS61p.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 88],
							data: [null,null],
							form: null,
							error: null
						});
					});
				}
			</script>
		
Xet Storage Details

Size:: 31.6 kB
Xet hash:: 56a68c20ca2def175f295471954d0988b98cf32da194071ec7ec646526d4c75f
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.