bitsandbytes/pr_1521/en/explanations/resources.html

<meta charset="utf-8" /><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Papers, related resources &amp; how to cite&quot;,&quot;local&quot;:&quot;papers-related-resources--how-to-cite&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression (Jun 2023)&quot;,&quot;local&quot;:&quot;spqr-a-sparse-quantized-representation-for-near-lossless-llm-weight-compression-jun-2023&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;QLoRA: Efficient Finetuning of Quantized LLMs (May 2023)&quot;,&quot;local&quot;:&quot;qlora-efficient-finetuning-of-quantized-llms-may-2023&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;The case for 4-bit precision: k-bit Inference Scaling Laws (Dec 2022)&quot;,&quot;local&quot;:&quot;the-case-for-4-bit-precision-k-bit-inference-scaling-laws-dec-2022&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale (Nov 2022)&quot;,&quot;local&quot;:&quot;llm-int8&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;8-bit Optimizers via Block-wise Quantization (Oct 2021)&quot;,&quot;local&quot;:&quot;8-bit-optimizers-via-block-wise-quantization-oct-2021&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}">
		<link href="/docs/bitsandbytes/pr_1521/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/entry/start.64267764.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/scheduler.852ec091.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/singletons.4904c6bc.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/index.268e315a.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/paths.7199c306.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/entry/app.42c0de65.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/index.28275fd3.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/nodes/0.2c7162e4.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/each.e59479a4.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/nodes/6.05e4e321.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/CodeBlock.c3366071.js">
		<link rel="modulepreload" href="/docs/bitsandbytes/pr_1521/en/_app/immutable/chunks/EditOnGithub.582011f0.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{&quot;title&quot;:&quot;Papers, related resources &amp; how to cite&quot;,&quot;local&quot;:&quot;papers-related-resources--how-to-cite&quot;,&quot;sections&quot;:[{&quot;title&quot;:&quot;SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression (Jun 2023)&quot;,&quot;local&quot;:&quot;spqr-a-sparse-quantized-representation-for-near-lossless-llm-weight-compression-jun-2023&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;QLoRA: Efficient Finetuning of Quantized LLMs (May 2023)&quot;,&quot;local&quot;:&quot;qlora-efficient-finetuning-of-quantized-llms-may-2023&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;The case for 4-bit precision: k-bit Inference Scaling Laws (Dec 2022)&quot;,&quot;local&quot;:&quot;the-case-for-4-bit-precision-k-bit-inference-scaling-laws-dec-2022&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale (Nov 2022)&quot;,&quot;local&quot;:&quot;llm-int8&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2},{&quot;title&quot;:&quot;8-bit Optimizers via Block-wise Quantization (Oct 2021)&quot;,&quot;local&quot;:&quot;8-bit-optimizers-via-block-wise-quantization-oct-2021&quot;,&quot;sections&quot;:[],&quot;depth&quot;:2}],&quot;depth&quot;:1}"><!-- HEAD_svelte-u9bgzb_END -->      <p></p>  <h1 class="relative group"><a id="papers-related-resources--how-to-cite" 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="#papers-related-resources--how-to-cite"><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>Papers, related resources &amp; how to cite</span></h1> <p data-svelte-h="svelte-1jp7sfc">The below academic work is ordered in reverse chronological order.</p>  <h2 class="relative group"><a id="spqr-a-sparse-quantized-representation-for-near-lossless-llm-weight-compression-jun-2023" 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="#spqr-a-sparse-quantized-representation-for-near-lossless-llm-weight-compression-jun-2023"><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>SpQR: A Sparse-Quantized Representation for Near-Lossless LLM Weight Compression (Jun 2023)</span></h2> <p data-svelte-h="svelte-6ej7zg">Authors: Tim Dettmers, Ruslan Svirschevski, Vage Egiazarian, Denis Kuznedelev, Elias Frantar, Saleh Ashkboos, Alexander Borzunov, Torsten Hoefler, Dan Alistarh</p> <ul data-svelte-h="svelte-1bk8yei"><li><a href="https://twitter.com/Tim_Dettmers/status/1666076553665744896" rel="nofollow">Twitter summary thread</a></li></ul> <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 -->@article{dettmers2023spqr,
  title={SpQR: A Sparse-Quantized Representation for Near-Lossless <span class="hljs-keyword">LLM </span>Weight Compression},
  author={Dettmers, Tim <span class="hljs-keyword">and </span>Svirschevski, Ruslan <span class="hljs-keyword">and </span>Egiazarian, Vage <span class="hljs-keyword">and </span>Kuznedelev, Denis <span class="hljs-keyword">and </span>Frantar, Elias <span class="hljs-keyword">and </span>Ashkboos, Saleh <span class="hljs-keyword">and </span><span class="hljs-keyword">Borzunov, </span>Alexander <span class="hljs-keyword">and </span>Hoefler, Torsten <span class="hljs-keyword">and </span>Alistarh, Dan},
  <span class="hljs-keyword">journal={arXiv </span>preprint arXiv:<span class="hljs-number">2306</span>.<span class="hljs-number">03078</span>},
  year={<span class="hljs-number">2023</span>}
}<!-- HTML_TAG_END --></pre></div>  <h2 class="relative group"><a id="qlora-efficient-finetuning-of-quantized-llms-may-2023" 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="#qlora-efficient-finetuning-of-quantized-llms-may-2023"><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>QLoRA: Efficient Finetuning of Quantized LLMs (May 2023)</span></h2> <p data-svelte-h="svelte-1m3ezda">Authors: Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, Luke Zettlemoyer</p> <ul data-svelte-h="svelte-95bqwh"><li><a href="https://www.youtube.com/watch?v=y9PHWGOa8HA&ab_channel=LondonMachineLearningMeetup" rel="nofollow">Video</a></li> <li><a href="https://twitter.com/Tim_Dettmers/status/1661379354507476994" rel="nofollow">Twitter summary thread</a></li></ul> <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 -->@article{dettmers2023qlora,
  title={Qlora: Efficient finetuning of quantized <span class="hljs-keyword">llms},
</span>  author={Dettmers, Tim <span class="hljs-keyword">and </span>Pagnoni, Artidoro <span class="hljs-keyword">and </span>Holtzman, Ari <span class="hljs-keyword">and </span>Zettlemoyer, Luke},
  <span class="hljs-keyword">journal={arXiv </span>preprint arXiv:<span class="hljs-number">2305</span>.<span class="hljs-number">14314</span>},
  year={<span class="hljs-number">2023</span>}
}<!-- HTML_TAG_END --></pre></div>  <h2 class="relative group"><a id="the-case-for-4-bit-precision-k-bit-inference-scaling-laws-dec-2022" 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="#the-case-for-4-bit-precision-k-bit-inference-scaling-laws-dec-2022"><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>The case for 4-bit precision: k-bit Inference Scaling Laws (Dec 2022)</span></h2> <p data-svelte-h="svelte-1ltttkj">Authors: Tim Dettmers, Luke Zettlemoyer</p> <ul data-svelte-h="svelte-1uuma5m"><li><a href="https://www.youtube.com/watch?v=odlQa6AE1gY&ab_channel=TheInsideView" rel="nofollow">Video</a></li> <li><a href="https://twitter.com/Tim_Dettmers/status/1605209171758284805" rel="nofollow">Twitter summary thread</a></li></ul> <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="language-xml">@inproceedings</span><span class="hljs-template-variable">{dettmers2023case,
  title={The case for 4-bit precision: k-bit inference scaling laws}</span><span class="language-xml">,
  author=</span><span class="hljs-template-variable">{Dettmers, Tim and Zettlemoyer, Luke}</span><span class="language-xml">,
  booktitle=</span><span class="hljs-template-variable">{International Conference on Machine Learning}</span><span class="language-xml">,
  pages=</span><span class="hljs-template-variable">{7750--7774}</span><span class="language-xml">,
  year=</span><span class="hljs-template-variable">{2023}</span><span class="language-xml">,
  organization=</span><span class="hljs-template-variable">{PMLR}</span><span class="language-xml">
}</span><!-- HTML_TAG_END --></pre></div>  <h2 class="relative group"><a id="llm-int8" 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="#llm-int8"><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>LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale (Nov 2022)</span></h2> <p data-svelte-h="svelte-1d05yau">Authors: Tim Dettmers, Mike Lewis, Younes Belkada, Luke Zettlemoyer</p> <ul data-svelte-h="svelte-1w2oslr"><li><a href="https://huggingface.co/blog/hf-bitsandbytes-integration" rel="nofollow">LLM.int8() Blog Post</a></li> <li><a href="https://timdettmers.com/2022/08/17/llm-int8-and-emergent-features/" rel="nofollow">LLM.int8() Emergent Features Blog Post</a></li> <li><a href="https://towardsdatascience.com/introduction-to-weight-quantization-2494701b9c0c" rel="nofollow">Introduction to Weight Quantization</a></li> <li><a href="https://twitter.com/Tim_Dettmers/status/1598351301942951937" rel="nofollow">Poster</a></li></ul> <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 -->@article{dettmers2022llm,
  title={<span class="hljs-keyword">Llm. </span>int8 (): <span class="hljs-number">8</span>-<span class="hljs-keyword">bit </span>matrix <span class="hljs-keyword">multiplication </span>for transformers <span class="hljs-built_in">at</span> <span class="hljs-keyword">scale},
</span>  author={Dettmers, Tim <span class="hljs-keyword">and </span>Lewis, Mike <span class="hljs-keyword">and </span><span class="hljs-keyword">Belkada, </span>Younes <span class="hljs-keyword">and </span>Zettlemoyer, Luke},
  <span class="hljs-keyword">journal={arXiv </span>preprint arXiv:<span class="hljs-number">2208</span>.<span class="hljs-number">07339</span>},
  year={<span class="hljs-number">2022</span>}
}<!-- HTML_TAG_END --></pre></div>  <h2 class="relative group"><a id="8-bit-optimizers-via-block-wise-quantization-oct-2021" 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="#8-bit-optimizers-via-block-wise-quantization-oct-2021"><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>8-bit Optimizers via Block-wise Quantization (Oct 2021)</span></h2> <p data-svelte-h="svelte-cvnv0k">Authors: Tim Dettmers, Mike Lewis, Sam Shleifer, Luke Zettlemoyer</p> <ul data-svelte-h="svelte-9u33gj"><li><a href="https://www.youtube.com/watch?v=IxrlHAJtqKE" rel="nofollow">Video</a></li> <li><a href="https://twitter.com/Tim_Dettmers/status/1446472128979562499" rel="nofollow">Twitter summary thread</a></li></ul> <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 -->@article{DBLP:journals<span class="hljs-regexp">/corr/</span>abs-<span class="hljs-number">2110</span>-<span class="hljs-number">02861</span>,
  author       = {Tim Dettmers and
                  Mike Lewis and
                  Sam Shleifer and
                  Luke Zettlemoyer},
  title        = {<span class="hljs-number">8</span>-bit Optimizers via Block-wise Quantization},
  journal      = {CoRR},
  volume       = {abs/<span class="hljs-number">2110.02861</span>},
  year         = {<span class="hljs-number">2021</span>},
  url          = {https:<span class="hljs-regexp">//</span>arxiv.org<span class="hljs-regexp">/abs/</span><span class="hljs-number">2110.02861</span>},
  eprinttype    = {arXiv},
  eprint       = {<span class="hljs-number">2110.02861</span>},
  timestamp    = {Thu, <span class="hljs-number">21</span> Oct <span class="hljs-number">2021</span> <span class="hljs-number">16</span>:<span class="hljs-number">20</span>:<span class="hljs-number">08</span> +<span class="hljs-number">0200</span>},
  biburl       = {https:<span class="hljs-regexp">//</span>dblp.org<span class="hljs-regexp">/rec/</span>journals<span class="hljs-regexp">/corr/</span>abs-<span class="hljs-number">2110</span>-<span class="hljs-number">02861</span>.bib},
  bibsource    = {dblp computer science bibliography, https:<span class="hljs-regexp">//</span>dblp.org}
}<!-- 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/bitsandbytes-foundation/bitsandbytes/blob/main/docs/source/explanations/resources.mdx" 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_1s46fht = {
						assets: "/docs/bitsandbytes/pr_1521/en",
						base: "/docs/bitsandbytes/pr_1521/en",
						env: {}
					};

					const element = document.currentScript.parentElement;

					const data = [null,null];

					Promise.all([
						import("/docs/bitsandbytes/pr_1521/en/_app/immutable/entry/start.64267764.js"),
						import("/docs/bitsandbytes/pr_1521/en/_app/immutable/entry/app.42c0de65.js")
					]).then(([kit, app]) => {
						kit.start(app, element, {
							node_ids: [0, 6],
							data,
							form: null,
							error: null
						});
					});
				}
			</script>