Hugging Face's logo

Buckets:
hf-doc-build
/
doc-dev

Files
xet
hf-doc-build/doc-dev / transformers /main /ko /quantization /gptq.html
rtrm's picture
rtrm
download
raw
27.2 kB
<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;GPTQ&quot;,&quot;local&quot;:&quot;gptq&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;ExLlama&quot;,&quot;local&quot;:&quot;exllama&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.9aa88961.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/scheduler.9bc65507.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/singletons.9eec45c3.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/index.3b203c72.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/paths.566078f7.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/entry/app.84fb67c3.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/index.707bf1b6.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/nodes/0.1c99376b.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/54.927308f5.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/Tip.c2ecdbf4.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/CodeBlock.54a9f38d.js">
 <link rel="modulepreload" href="/docs/transformers/main/ko/_app/immutable/chunks/EditOnGithub.922df6ba.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;GPTQ&quot;,&quot;local&quot;:&quot;gptq&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;ExLlama&quot;,&quot;local&quot;:&quot;exllama&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="gptq" 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="#gptq"><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>GPTQ</span></h1> <div class="course-tip bg-gradient-to-br dark:bg-gradient-to-r before:border-green-500 dark:before:border-green-800 from-green-50 dark:from-gray-900 to-white dark:to-gray-950 border border-green-50 text-green-700 dark:text-gray-400"><p data-svelte-h="svelte-1da9pno">PEFT를 활용한 GPTQ 양자화를 사용해보시려면 이 <a href="https://colab.research.google.com/drive/1_TIrmuKOFhuRRiTWN94iLKUFu6ZX4ceb" rel="nofollow">노트북</a>을 참고하시고, 자세한 내용은 이 <a href="https://huggingface.co/blog/gptq-integration" rel="nofollow">블로그 게시물</a>에서 확인하세요!</p></div> <p data-svelte-h="svelte-1vdg1r0"><a href="https://github.com/PanQiWei/AutoGPTQ" rel="nofollow">AutoGPTQ</a> 라이브러리는 GPTQ 알고리즘을 구현합니다. 이는 훈련 후 양자화 기법으로, 가중치 행렬의 각 행을 독립적으로 양자화하여 오차를 최소화하는 가중치 버전을 찾습니다. 이 가중치는 int4로 양자화되지만, 추론 중에는 실시간으로 fp16으로 복원됩니다. 이는 int4 가중치가 GPU의 전역 메모리 대신 결합된 커널에서 역양자화되기 때문에 메모리 사용량을 4배 절약할 수 있으며, 더 낮은 비트 너비를 사용함으로써 통신 시간이 줄어들어 추론 속도가 빨라질 것으로 기대할 수 있습니다.</p> <p data-svelte-h="svelte-1lpb91s">시작하기 전에 다음 라이브러리들이 설치되어 있는지 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=""><!-- HTML_TAG_START -->pip install auto-gptq
pip install --upgrade accelerate optimum transformers<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-i1p7pi">모델을 양자화하려면(현재 텍스트 모델만 지원됨) <code>GPTQConfig</code> 클래스를 생성하고 양자화할 비트 수, 양자화를 위한 가중치 교정 데이터셋, 그리고 데이터셋을 준비하기 위한 토크나이저를 설정해야 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, AutoTokenizer, GPTQConfig
model_id = <span class="hljs-string">&quot;facebook/opt-125m&quot;</span>
tokenizer = AutoTokenizer.from_pretrained(model_id)
gptq_config = GPTQConfig(bits=<span class="hljs-number">4</span>, dataset=<span class="hljs-string">&quot;c4&quot;</span>, tokenizer=tokenizer)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-16sjond">자신의 데이터셋을 문자열 리스트 형태로 전달할 수도 있지만, GPTQ 논문에서 사용한 동일한 데이터셋을 사용하는 것을 강력히 권장합니다.</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 -->dataset = [<span class="hljs-string">&quot;auto-gptq is an easy-to-use model quantization library with user-friendly apis, based on GPTQ algorithm.&quot;</span>]
gptq_config = GPTQConfig(bits=<span class="hljs-number">4</span>, dataset=dataset, tokenizer=tokenizer)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1mjb5cu">양자화할 모델을 로드하고 <code>gptq_config</code>을 <code>from_pretrained()</code> 메소드에 전달하세요. 모델을 메모리에 맞추기 위해 <code>device_map=&quot;auto&quot;</code>를 설정하여 모델을 자동으로 CPU로 오프로드하고, 양자화를 위해 모델 모듈이 CPU와 GPU 간에 이동할 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=""><!-- HTML_TAG_START -->quantized_model = AutoModelForCausalLM.from_pretrained(model_id, device_map=<span class="hljs-string">&quot;auto&quot;</span>, quantization_config=gptq_config)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-lj3tj5">데이터셋이 너무 커서 메모리가 부족한 경우를 대비한 디스크 오프로드는 현재 지원하지 않고 있습니다. 이럴 때는 <code>max_memory</code> 매개변수를 사용하여 디바이스(GPU 및 CPU)에서 사용할 메모리 양을 할당해 보세요:</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 -->quantized_model = AutoModelForCausalLM.from_pretrained(model_id, device_map=<span class="hljs-string">&quot;auto&quot;</span>, max_memory={<span class="hljs-number">0</span>: <span class="hljs-string">&quot;30GiB&quot;</span>, <span class="hljs-number">1</span>: <span class="hljs-string">&quot;46GiB&quot;</span>, <span class="hljs-string">&quot;cpu&quot;</span>: <span class="hljs-string">&quot;30GiB&quot;</span>}, quantization_config=gptq_config)<!-- HTML_TAG_END --></pre></div> <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-1yg57hz">하드웨어와 모델 매개변수량에 따라 모델을 처음부터 양자화하는 데 드는 시간이 서로 다를 수 있습니다. 예를 들어, 무료 등급의 Google Colab GPU로 비교적 가벼운 <a href="https://huggingface.co/facebook/opt-350m" rel="nofollow">facebook/opt-350m</a> 모델을 양자화하는 데 약 5분이 걸리지만, NVIDIA A100으로 175B에 달하는 매개변수를 가진 모델을 양자화하는 데는 약 4시간에 달하는 시간이 걸릴 수 있습니다. 모델을 양자화하기 전에, Hub에서 해당 모델의 GPTQ 양자화 버전이 이미 존재하는지 확인하는 것이 좋습니다.</p></div> <p data-svelte-h="svelte-68znt5">모델이 양자화되면, 모델과 토크나이저를 Hub에 푸시하여 쉽게 공유하고 접근할 수 있습니다. <code>GPTQConfig</code>를 저장하기 위해 <code>push_to_hub()</code> 메소드를 사용하세요:</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 -->quantized_model.push_to_hub(<span class="hljs-string">&quot;opt-125m-gptq&quot;</span>)
tokenizer.push_to_hub(<span class="hljs-string">&quot;opt-125m-gptq&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-1ou4phu">양자화된 모델을 로컬에 저장하려면 <code>save_pretrained()</code> 메소드를 사용할 수 있습니다. 모델이 <code>device_map</code> 매개변수로 양자화되었을 경우, 저장하기 전에 전체 모델을 GPU나 CPU로 이동해야 합니다. 예를 들어, 모델을 CPU에 저장하려면 다음과 같이 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=""><!-- HTML_TAG_START -->quantized_model.save_pretrained(<span class="hljs-string">&quot;opt-125m-gptq&quot;</span>)
tokenizer.save_pretrained(<span class="hljs-string">&quot;opt-125m-gptq&quot;</span>)
<span class="hljs-comment"># device_map이 설정된 상태에서 양자화된 경우</span>
quantized_model.to(<span class="hljs-string">&quot;cpu&quot;</span>)
quantized_model.save_pretrained(<span class="hljs-string">&quot;opt-125m-gptq&quot;</span>)<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-a0kib9">양자화된 모델을 다시 로드하려면 <code>from_pretrained()</code> 메소드를 사용하고, <code>device_map=&quot;auto&quot;</code>를 설정하여 모든 사용 가능한 GPU에 모델을 자동으로 분산시켜 더 많은 메모리를 사용하지 않으면서 모델을 더 빠르게 로드할 수 있습니다.</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> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM
model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">&quot;{your_username}/opt-125m-gptq&quot;</span>, device_map=<span class="hljs-string">&quot;auto&quot;</span>)<!-- HTML_TAG_END --></pre></div> <h2 class="relative group"><a id="exllama" 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="#exllama"><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>ExLlama</span></h2> <p data-svelte-h="svelte-1s3k1d1"><a href="https://github.com/turboderp/exllama" rel="nofollow">ExLlama</a>은 <a href="model_doc/llama">Llama</a> 모델의 Python/C++/CUDA 구현체로, 4비트 GPTQ 가중치를 사용하여 더 빠른 추론을 위해 설계되었습니다(이 <a href="https://github.com/huggingface/optimum/tree/main/tests/benchmark#gptq-benchmark" rel="nofollow">벤치마크</a>를 참고하세요). [‘GPTQConfig’] 객체를 생성할 때 ExLlama 커널이 기본적으로 활성화됩니다. 추론 속도를 더욱 높이기 위해, <code>exllama_config</code> 매개변수를 구성하여 <a href="https://github.com/turboderp/exllamav2" rel="nofollow">ExLlamaV2</a> 커널을 사용할 수 있습니다:</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">import</span> torch
<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, GPTQConfig
gptq_config = GPTQConfig(bits=<span class="hljs-number">4</span>, exllama_config={<span class="hljs-string">&quot;version&quot;</span>:<span class="hljs-number">2</span>})
model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">&quot;{your_username}/opt-125m-gptq&quot;</span>, device_map=<span class="hljs-string">&quot;auto&quot;</span>, quantization_config=gptq_config)<!-- HTML_TAG_END --></pre></div> <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-rq1jxq">4비트 모델만 지원되며, 양자화된 모델을 PEFT로 미세 조정하는 경우 ExLlama 커널을 비활성화할 것을 권장합니다.</p></div> <p data-svelte-h="svelte-wx4irt">ExLlama 커널은 전체 모델이 GPU에 있을 때만 지원됩니다. AutoGPTQ(버전 0.4.2 이상)로 CPU에서 추론을 수행하는 경우 ExLlama 커널을 비활성화해야 합니다. 이를 위해 config.json 파일의 양자화 설정에서 ExLlama 커널과 관련된 속성을 덮어써야 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=""><!-- HTML_TAG_START --><span class="hljs-keyword">import</span> torch
<span class="hljs-keyword">from</span> transformers <span class="hljs-keyword">import</span> AutoModelForCausalLM, GPTQConfig
gptq_config = GPTQConfig(bits=<span class="hljs-number">4</span>, use_exllama=<span class="hljs-literal">False</span>)
model = AutoModelForCausalLM.from_pretrained(<span class="hljs-string">&quot;{your_username}/opt-125m-gptq&quot;</span>, device_map=<span class="hljs-string">&quot;cpu&quot;</span>, quantization_config=gptq_config)<!-- 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/quantization/gptq.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_1hrx8 = {
 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.9aa88961.js"),
 import("/docs/transformers/main/ko/_app/immutable/entry/app.84fb67c3.js")
 ]).then(([kit, app]) => {
 kit.start(app, element, {
 node_ids: [0, 54],
 data,
 form: null,
 error: null
 });
 });
 }
 </script>

Xet Storage Details

Size:
27.2 kB
·
Xet hash:
216b689d9d6e53578e2f0789018a2712e5ba3b9b6e92d9a51b1c799c3a6c378a

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