transformers/main/ko/how_to_hack_models.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;모델 구성 요소 맞춤 설정하기&quot;,&quot;local&quot;:&quot;customizing-model-components&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;어텐션 클래스&quot;,&quot;local&quot;:&quot;attention-class&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;LoRA&quot;,&quot;local&quot;:&quot;lora&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
		<link href="/docs/transformers/main/ko/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/entry/start.3df1c19e.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/scheduler.53228c21.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/singletons.83674bae.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/index.e93d0901.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/paths.aee36068.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/entry/app.5226fb4b.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/preload-helper.cb103237.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/index.3db2ce32.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/nodes/0.46820ae2.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/each.e59479a4.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/nodes/21.0e78f159.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/CopyLLMTxtMenu.1327b590.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/MermaidChart.svelte_svelte_type_style_lang.49b88d99.js">
		<link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/CodeBlock.ada04ea6.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;모델 구성 요소 맞춤 설정하기&quot;,&quot;local&quot;:&quot;customizing-model-components&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;어텐션 클래스&quot;,&quot;local&quot;:&quot;attention-class&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;LoRA&quot;,&quot;local&quot;:&quot;lora&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="customizing-model-components" 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="#customizing-model-components"><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>모델 구성 요소 맞춤 설정하기</span></h1> <p data-svelte-h="svelte-bnl9yr">모델을 완전히 새로 작성하는 대신 구성 요소를 수정하여 모델을 맞춤 설정하는 방법이 있습니다. 이 방법으로 모델을 특정 사용 사례에 맞게 모델을 조정할 수 있습니다. 예를 들어, 새로운 레이어를 추가하거나 아키텍처의 어텐션 메커니즘을 최적화할 수 있습니다. 이러한 맞춤 설정은 트랜스포머 모델에 직접 적용되므로, <a href="/docs/transformers/main/ko/main_classes/trainer#transformers.Trainer">Trainer</a>, <a href="/docs/transformers/main/ko/main_classes/model#transformers.PreTrainedModel">PreTrainedModel</a> 및 <a href="https://huggingface.co/docs/peft/en/index" rel="nofollow">PEFT</a> 라이브러리와 같은 기능을 계속 사용할 수 있습니다.</p> <p data-svelte-h="svelte-s8t39s">이 가이드에서는 모델의 어텐션 메커니즘을 맞춤 설정하여 <a href="https://huggingface.co/docs/peft/conceptual_guides/adapter#low-rank-adaptation-lora" rel="nofollow">Low-Rank Adaptation (LoRA)</a>를 적용하는 방법을 설명합니다.</p> <blockquote class="tip"><p data-svelte-h="svelte-172i5m1">모델 코드를 반복적으로 수정하고 개발할 때 <a href="https://github.com/huggingface/transformers/blob/9985d06add07a4cc691dc54a7e34f54205c04d40/src/transformers/utils/import_utils.py#L2286" rel="nofollow">clear_import_cache</a> 유틸리티가 매우 유용합니다. 이 기능은 캐시된 모든 트랜스포머 모듈을 제거하여 Python이 환경을 재시작하지 않고도 수정된 코드를 다시 가져올 수 있도록 합니다.</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> AutoModel
<span class="hljs-keyword">from</span> transformers.utils.import_utils <span class="hljs-keyword">import</span> clear_import_cache

model = AutoModel.from_pretrained(<span class="hljs-string">&quot;bert-base-uncased&quot;</span>)
<span class="hljs-comment"># 모델 코드 수정</span>
<span class="hljs-comment"># 캐시를 지워 수정된 코드를 다시 가져오기</span>
clear_import_cache()
<span class="hljs-comment"># 업데이트된 코드를 사용하기 위해 다시 가져오기</span>
model = AutoModel.from_pretrained(<span class="hljs-string">&quot;bert-base-uncased&quot;</span>)<!-- HTML_TAG_END --></pre></div></blockquote>  <h2 class="relative group"><a id="attention-class" 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="#attention-class"><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>어텐션 클래스</span></h2> <p data-svelte-h="svelte-78zvml"><a href="./model_doc/sam">Segment Anything</a>은 이미지 분할 모델로, 어텐션 메커니즘에서 query-key-value(<code>qkv</code>) 프로젝션을 결합합니다. 학습 가능한 파라미터 수와 연산 부담을 줄이기 위해 <code>qkv</code> 프로젝션에 LoRA를 적용할 수 있습니다. 이를 위해서는 <code>qkv</code> 프로젝션을 분리하여 <code>q</code>와 <code>v</code>에 LoRA를 개별적으로 적용해야 합니다.</p> <ol data-svelte-h="svelte-txqnd5"><li>원래의 <code>SamVisionAttention</code> 클래스를 상속하여 <code>SamVisionAttentionSplit</code>이라는 사용자 정의 어텐션 클래스를 만듭니다. <code>__init__</code>에서 결합된 <code>qkv</code>를 삭제하고, <code>q</code>, <code>k</code>, <code>v</code>를 위한 개별 선형 레이어를 생성합니다.</li></ol> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5   text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class="language-py "><!-- HTML_TAG_START --><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">from</span> transformers.models.sam.modeling_sam <span class="hljs-keyword">import</span> SamVisionAttention

<span class="hljs-keyword">class</span> <span class="hljs-title class_">SamVisionAttentionSplit</span>(SamVisionAttention, nn.Module):
    <span class="hljs-keyword">def</span> <span class="hljs-title function_">__init__</span>(<span class="hljs-params">self, config, window_size</span>):
        <span class="hljs-built_in">super</span>().__init__(config, window_size)
        <span class="hljs-comment"># 결합된 qkv 제거</span>
        <span class="hljs-keyword">del</span> self.qkv
        <span class="hljs-comment"># q, k, v 개별 프로젝션 생성</span>
        self.q = nn.Linear(config.hidden_size, config.hidden_size, bias=config.qkv_bias)
        self.k = nn.Linear(config.hidden_size, config.hidden_size, bias=config.qkv_bias)
        self.v = nn.Linear(config.hidden_size, config.hidden_size, bias=config.qkv_bias)
        self._register_load_state_dict_pre_hook(self.split_q_k_v_load_hook)<!-- HTML_TAG_END --></pre></div> <ol start="2" data-svelte-h="svelte-1hc0eq2"><li><code>_split_qkv_load_hook</code> 함수는 모델을 가져올 때, 사전 훈련된 <code>qkv</code> 가중치를 <code>q</code>, <code>k</code>, <code>v</code>로 분리하여 사전 훈련된 모델과의 호환성을 보장합니다.</li></ol> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5   text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class="language-py "><!-- HTML_TAG_START -->    <span class="hljs-keyword">def</span> <span class="hljs-title function_">split_q_k_v_load_hook</span>(<span class="hljs-params">self, state_dict, prefix, *args</span>):
        keys_to_delete = []
        <span class="hljs-keyword">for</span> key <span class="hljs-keyword">in</span> <span class="hljs-built_in">list</span>(state_dict.keys()):
            <span class="hljs-keyword">if</span> <span class="hljs-string">&quot;qkv.&quot;</span> <span class="hljs-keyword">in</span> key:
                <span class="hljs-comment"># 결합된 프로젝션에서 q, k, v 분리</span>
                q, k, v = state_dict[key].chunk(<span class="hljs-number">3</span>, dim=<span class="hljs-number">0</span>)
                <span class="hljs-comment"># 개별 q, k, v 프로젝션으로 대체</span>
                state_dict[key.replace(<span class="hljs-string">&quot;qkv.&quot;</span>, <span class="hljs-string">&quot;q.&quot;</span>)] = q
                state_dict[key.replace(<span class="hljs-string">&quot;qkv.&quot;</span>, <span class="hljs-string">&quot;k.&quot;</span>)] = k
                state_dict[key.replace(<span class="hljs-string">&quot;qkv.&quot;</span>, <span class="hljs-string">&quot;v.&quot;</span>)] = v
                <span class="hljs-comment"># 기존 qkv 키를 삭제 대상으로 표시</span>
                keys_to_delete.append(key)
        
        <span class="hljs-comment"># 기존 qkv 키 제거</span>
        <span class="hljs-keyword">for</span> key <span class="hljs-keyword">in</span> keys_to_delete:
            <span class="hljs-keyword">del</span> state_dict[key]<!-- HTML_TAG_END --></pre></div> <ol start="3" data-svelte-h="svelte-1fplex1"><li><code>forward</code> 단계에서 <code>q</code>, <code>k</code>, <code>v</code>는 개별적으로 계산되며, 어텐션 메커니즘의 나머지 부분은 동일하게 유지됩니다.</li></ol> <div class="code-block relative "><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5   text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class="language-py "><!-- HTML_TAG_START -->    <span class="hljs-keyword">def</span> <span class="hljs-title function_">forward</span>(<span class="hljs-params">self, hidden_states: torch.Tensor, output_attentions=<span class="hljs-literal">False</span></span>) -&gt; torch.Tensor:
        batch_size, height, width, _ = hidden_states.shape
        qkv_shapes = (batch_size *  self.num_attention_heads,  height * width, -<span class="hljs-number">1</span>)
        query = self.q(hidden_states).reshape((batch_size,  height * width,self.num_attention_heads, -<span class="hljs-number">1</span>)).permute(<span class="hljs-number">0</span>,<span class="hljs-number">2</span>,<span class="hljs-number">1</span>,<span class="hljs-number">3</span>).reshape(qkv_shapes)
        key = self.k(hidden_states).reshape((batch_size,  height * width,self.num_attention_heads, -<span class="hljs-number">1</span>)).permute(<span class="hljs-number">0</span>,<span class="hljs-number">2</span>,<span class="hljs-number">1</span>,<span class="hljs-number">3</span>).reshape(qkv_shapes)
        value = self.v(hidden_states).reshape((batch_size,  height * width,self.num_attention_heads, -<span class="hljs-number">1</span>)).permute(<span class="hljs-number">0</span>,<span class="hljs-number">2</span>,<span class="hljs-number">1</span>,<span class="hljs-number">3</span>).reshape(qkv_shapes)

        attn_weights = (query * self.scale) @ key.transpose(-<span class="hljs-number">2</span>, -<span class="hljs-number">1</span>)

        attn_weights = torch.nn.functional.softmax(attn_weights, dtype=torch.float32, dim=-<span class="hljs-number">1</span>).to(query.dtype)
        attn_probs = nn.functional.dropout(attn_weights, p=self.dropout, training=self.training)
        attn_output = (attn_probs @ value).reshape(batch_size, self.num_attention_heads, height, width, -<span class="hljs-number">1</span>)
        attn_output = attn_output.permute(<span class="hljs-number">0</span>, <span class="hljs-number">2</span>, <span class="hljs-number">3</span>, <span class="hljs-number">1</span>, <span class="hljs-number">4</span>).reshape(batch_size, height, width, -<span class="hljs-number">1</span>)
        attn_output = self.proj(attn_output)

        <span class="hljs-keyword">if</span> output_attentions:
            outputs = (attn_output, attn_weights)
        <span class="hljs-keyword">else</span>:
            outputs = (attn_output, <span class="hljs-literal">None</span>)
        <span class="hljs-keyword">return</span> outputs<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1nynq3b">사용자 정의 <code>SamVisionAttentionSplit</code> 클래스를 원본 모델의 <code>SamVisionAttention</code> 모듈에 할당하여 교체합니다. 모델 내 모든 <code>SamVisionAttention</code> 인스턴스는 분리된 어텐션 버전으로 대체됩니다.</p> <p data-svelte-h="svelte-9ri3ij"><a href="/docs/transformers/main/ko/main_classes/model#transformers.PreTrainedModel.from_pretrained">from_pretrained()</a>로 모델을 가져오세요.</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> SamModel

<span class="hljs-comment"># 사전 훈련된 SAM 모델 가져오기</span>
model = SamModel.from_pretrained(<span class="hljs-string">&quot;facebook/sam-vit-base&quot;</span>)

<span class="hljs-comment"># 비전-인코더 모듈에서 어텐션 클래스 교체</span>
<span class="hljs-keyword">for</span> layer <span class="hljs-keyword">in</span> model.vision_encoder.layers:
    <span class="hljs-keyword">if</span> <span class="hljs-built_in">hasattr</span>(layer, <span class="hljs-string">&quot;attn&quot;</span>):
        layer.attn = SamVisionAttentionSplit(model.config.vision_config, model.config.vision_config.window_size)<!-- HTML_TAG_END --></pre></div>  <h2 class="relative group"><a id="lora" 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="#lora"><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>LoRA</span></h2> <p data-svelte-h="svelte-16e655">분리된 <code>q</code>, <code>k</code>, <code>v</code> 프로젝션을 사용할 때 , <code>q</code>와 <code>v</code>에 LoRA를 적용합니다.</p> <p data-svelte-h="svelte-urjfm3"><a href="https://huggingface.co/docs/peft/package_reference/config#peft.PeftConfig" rel="nofollow">LoraConfig</a>를 생성하고, 랭크 <code>r</code>, <code>lora_alpha</code>, <code>lora_dropout</code>, <code>task_type</code>, 그리고 가장 중요한 적용될 모듈을 지정합니다.</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> peft <span class="hljs-keyword">import</span> LoraConfig, get_peft_model

config = LoraConfig(
    r=<span class="hljs-number">16</span>,
    lora_alpha=<span class="hljs-number">32</span>,
    <span class="hljs-comment"># q와 v에 LoRA 적용</span>
    target_modules=[<span class="hljs-string">&quot;q&quot;</span>, <span class="hljs-string">&quot;v&quot;</span>],
    lora_dropout=<span class="hljs-number">0.1</span>,
    task_type=<span class="hljs-string">&quot;FEATURE_EXTRACTION&quot;</span>
)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1w0ksm9">모델과 <a href="https://huggingface.co/docs/peft/package_reference/config#peft.PeftConfig" rel="nofollow">LoraConfig</a>를 <a href="https://huggingface.co/docs/peft/package_reference/peft_model#peft.get_peft_model" rel="nofollow">get_peft_model</a>에 전달하여 모델에 LoRA를 적용합니다.</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 -->model = get_peft_model(model, config)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-14kdo1h"><a href="https://huggingface.co/docs/peft/package_reference/peft_model#peft.PeftMixedModel.print_trainable_parameters" rel="nofollow">print_trainable_parameters</a>를 호출하여 전체 파라미터 수 대비 훈련되는 파라미터 수를 확인하세요.</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 -->model.print_trainable_parameters()
<span class="hljs-string">&quot;trainable params: 589,824 || all params: 94,274,096 || trainable%: 0.6256&quot;</span><!-- HTML_TAG_END --></pre></div> <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/ko/how_to_hack_models.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_1vzb9oe = {
						assets: "/docs/transformers/main/ko",
						base: "/docs/transformers/main/ko",
						env: {}
					};

					const element = document.currentScript.parentElement;

					const data = [null,null];

					Promise.all([
						import("/docs/transformers/main/ko/_app/immutable/entry/start.3df1c19e.js"),
						import("/docs/transformers/main/ko/_app/immutable/entry/app.5226fb4b.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 21],
							data,
							form: null,
							error: null
						});
					});
				}
			</script>