Buckets:
| <meta charset="utf-8" /><meta name="hf:doc:metadata" content="{"title":"Installation Guide","local":"installation-guide","sections":[{"title":"Table of Contents","local":"table-of-contents","sections":[],"depth":2},{"title":"CUDA","local":"cuda","sections":[{"title":"Supported CUDA Configurations","local":"cuda-pip","sections":[],"depth":3},{"title":"pip install pre-built wheel from latest main commit","local":"pip-install-pre-built-wheel-from-latest-main-commit","sections":[],"depth":3},{"title":"Compile from source","local":"cuda-compile","sections":[],"depth":3},{"title":"PyTorch CUDA versions","local":"pytorch-cuda-versions","sections":[],"depth":3}],"depth":2},{"title":"Multi-backend Support (Alpha Release)","local":"multi-backend","sections":[{"title":"Supported Backends","local":"multi-backend-supported-backends","sections":[],"depth":3},{"title":"Pre-requisites","local":"multi-backend-pre-requisites","sections":[],"depth":3},{"title":"Installation","local":"installation","sections":[{"title":"Pre-built Wheel Installation (recommended)","local":"multi-backend-pip","sections":[],"depth":4},{"title":"Compile from Source","local":"multi-backend-compile","sections":[],"depth":4},{"title":"AMD GPU","local":"amd-gpu","sections":[],"depth":4},{"title":"Intel CPU","local":"intel-cpu","sections":[],"depth":4},{"title":"Apple Silicon","local":"apple-silicon","sections":[],"depth":4}],"depth":3}],"depth":2}],"depth":1}"> | |
| <link href="/docs/bitsandbytes/pr_1137/en/_app/immutable/assets/0.e3b0c442.css" rel="modulepreload"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/entry/start.435ebb3a.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/scheduler.852ec091.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/singletons.c3371655.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/index.268e315a.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/paths.5769d97d.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/entry/app.6b2aadb8.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/index.28275fd3.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/nodes/0.fece9a53.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/each.e59479a4.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/nodes/10.05522dc1.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/Tip.9f398c59.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/CodeBlock.c3366071.js"> | |
| <link rel="modulepreload" href="/docs/bitsandbytes/pr_1137/en/_app/immutable/chunks/EditOnGithub.582011f0.js"><!-- HEAD_svelte-u9bgzb_START --><meta name="hf:doc:metadata" content="{"title":"Installation Guide","local":"installation-guide","sections":[{"title":"Table of Contents","local":"table-of-contents","sections":[],"depth":2},{"title":"CUDA","local":"cuda","sections":[{"title":"Supported CUDA Configurations","local":"cuda-pip","sections":[],"depth":3},{"title":"pip install pre-built wheel from latest main commit","local":"pip-install-pre-built-wheel-from-latest-main-commit","sections":[],"depth":3},{"title":"Compile from source","local":"cuda-compile","sections":[],"depth":3},{"title":"PyTorch CUDA versions","local":"pytorch-cuda-versions","sections":[],"depth":3}],"depth":2},{"title":"Multi-backend Support (Alpha Release)","local":"multi-backend","sections":[{"title":"Supported Backends","local":"multi-backend-supported-backends","sections":[],"depth":3},{"title":"Pre-requisites","local":"multi-backend-pre-requisites","sections":[],"depth":3},{"title":"Installation","local":"installation","sections":[{"title":"Pre-built Wheel Installation (recommended)","local":"multi-backend-pip","sections":[],"depth":4},{"title":"Compile from Source","local":"multi-backend-compile","sections":[],"depth":4},{"title":"AMD GPU","local":"amd-gpu","sections":[],"depth":4},{"title":"Intel CPU","local":"intel-cpu","sections":[],"depth":4},{"title":"Apple Silicon","local":"apple-silicon","sections":[],"depth":4}],"depth":3}],"depth":2}],"depth":1}"><!-- HEAD_svelte-u9bgzb_END --> <p></p> <h1 class="relative group"><a id="installation-guide" 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="#installation-guide"><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>Installation Guide</span></h1> <p data-svelte-h="svelte-1jlm1j8">Welcome to the installation guide for the <code>bitsandbytes</code> library! This document provides step-by-step instructions to install <code>bitsandbytes</code> across various platforms and hardware configurations. The library primarily supports CUDA-based GPUs, but the team is actively working on enabling support for additional backends like AMD ROCm, Intel, and Apple Silicon.</p> <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-cjext7">For a high-level overview of backend support and compatibility, see the <a href="#multi-backend">Multi-backend Support</a> section.</p></div> <h2 class="relative group"><a id="table-of-contents" 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="#table-of-contents"><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>Table of Contents</span></h2> <ul data-svelte-h="svelte-152x96h"><li><a href="#cuda">CUDA</a><ul><li><a href="#cuda-pip">Installation via PyPI</a></li> <li><a href="#cuda-compile">Compile from Source</a></li></ul></li> <li><a href="#multi-backend">Multi-backend Support (Alpha Release)</a><ul><li><a href="#multi-backend-supported-backends">Supported Backends</a></li> <li><a href="#multi-backend-pre-requisites">Pre-requisites</a></li> <li><a href="#multi-backend-pip">Installation</a></li> <li><a href="#multi-backend-compile">Compile from Source</a></li></ul></li> <li><a href="#pytorch-cuda-versions">PyTorch CUDA Versions</a></li></ul> <h2 class="relative group"><a id="cuda" 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="#cuda"><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>CUDA</span></h2> <p data-svelte-h="svelte-1pojf9z"><code>bitsandbytes</code> is currently only supported on CUDA GPUs for CUDA versions <strong>11.0 - 12.5</strong>. However, there’s an ongoing multi-backend effort under development, which is currently in alpha. If you’re interested in providing feedback or testing, check out <a href="#multi-backend">the multi-backend section below</a>.</p> <h3 class="relative group"><a id="cuda-pip" 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="#cuda-pip"><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>Supported CUDA Configurations</span></h3> <p data-svelte-h="svelte-8ary7j">The latest version of <code>bitsandbytes</code> builds on the following configurations:</p> <table data-svelte-h="svelte-x9atc8"><thead><tr><th><strong>OS</strong></th> <th><strong>CUDA Version</strong></th> <th><strong>Compiler</strong></th></tr></thead> <tbody><tr><td><strong>Linux</strong></td> <td>11.7 - 12.3</td> <td>GCC 11.4</td></tr> <tr><td></td> <td>12.4+</td> <td>GCC 13.2</td></tr> <tr><td><strong>Windows</strong></td> <td>11.7 - 12.4</td> <td>MSVC 19.38+ (VS2022)</td></tr></tbody></table> <p data-svelte-h="svelte-10cczy">For Linux systems, ensure your hardware meets the following requirements:</p> <table data-svelte-h="svelte-ube75n"><thead><tr><th><strong>Feature</strong></th> <th><strong>Hardware Requirement</strong></th></tr></thead> <tbody><tr><td>LLM.int8()</td> <td>NVIDIA Turing (RTX 20 series, T4) or Ampere (RTX 30 series, A4-A100) GPUs</td></tr> <tr><td>8-bit optimizers/quantization</td> <td>NVIDIA Kepler (GTX 780 or newer)</td></tr></tbody></table> <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-1dkg4ti"><code>bitsandbytes >= 0.39.1</code> no longer includes Kepler binaries in pip installations. This requires <a href="#cuda-compile">manual compilation using</a> the <code>cuda11x_nomatmul_kepler</code> configuration.</p></div> <p data-svelte-h="svelte-16utcew">To install from PyPI.</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 bitsandbytes<!-- HTML_TAG_END --></pre></div> <h3 class="relative group"><a id="pip-install-pre-built-wheel-from-latest-main-commit" 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="#pip-install-pre-built-wheel-from-latest-main-commit"><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>pip install pre-built wheel from latest main commit</span></h3> <p data-svelte-h="svelte-wzqh43">If you would like to use new feature even before they are officially released and help us test them, feel free to install the wheel directly from our CI (<em>the wheel links will remain stable!</em>):</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">Linux </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Windows </div></div> <div class="language-select"><div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-comment"># Note, if you don't want to reinstall BNBs dependencies, append the `--no-deps` flag!</span> | |
| <span class="hljs-attribute">pip</span> install --force-reinstall 'https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_main/bitsandbytes-<span class="hljs-number">0</span>.<span class="hljs-number">44</span>.<span class="hljs-number">2</span>.dev0-py3-none-manylinux_2_24_x86_64.whl'<!-- HTML_TAG_END --></pre></div> </div> <h3 class="relative group"><a id="cuda-compile" 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="#cuda-compile"><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>Compile from source</span></h3> <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-4d0ae7">Don’t hesitate to compile from source! The process is pretty straight forward and resilient. This might be needed for older CUDA versions or other less common configurations, which we don’t support out of the box due to package size.</p></div> <p data-svelte-h="svelte-1ovl1gx">For Linux and Windows systems, compiling from source allows you to customize the build configurations. See below for detailed platform-specific instructions (see the <code>CMakeLists.txt</code> if you want to check the specifics and explore some additional options):</p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">Linux </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Windows </div></div> <div class="language-select"><p data-svelte-h="svelte-zs8fpy">To compile from source, you need CMake >= <strong>3.22.1</strong> and Python >= <strong>3.8</strong> installed. Make sure you have a compiler installed to compile C++ (<code>gcc</code>, <code>make</code>, headers, etc.).</p> <p data-svelte-h="svelte-noyxvz">For example, to install a compiler and CMake on Ubuntu:</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 -->apt-get install -y build-essential cmake<!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-gmjw5q">You should also install CUDA Toolkit by following the <a href="https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html" rel="nofollow">NVIDIA CUDA Installation Guide for Linux</a> guide from NVIDIA. The current expected CUDA Toolkit version is <strong>11.1+</strong> and it is recommended to install <strong>GCC >= 7.3</strong> and required to have at least <strong>GCC >= 6</strong>.</p> <p data-svelte-h="svelte-1agbdv5">Refer to the following table if you’re using another CUDA Toolkit version.</p> <table data-svelte-h="svelte-1nj1xe4"><thead><tr><th>CUDA Toolkit</th> <th>GCC</th></tr></thead> <tbody><tr><td>>= 11.4.1</td> <td>>= 11</td></tr> <tr><td>>= 12.0</td> <td>>= 12</td></tr> <tr><td>>= 12.4</td> <td>>= 13</td></tr></tbody></table> <p data-svelte-h="svelte-1fitvy1">Now to install the bitsandbytes package from source, run the following commands:</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 -->git <span class="hljs-built_in">clone</span> https://github.com/bitsandbytes-foundation/bitsandbytes.git && <span class="hljs-built_in">cd</span> bitsandbytes/ | |
| pip install -r requirements-dev.txt | |
| cmake -DCOMPUTE_BACKEND=cuda -S . | |
| make | |
| pip install -e . <span class="hljs-comment"># `-e` for "editable" install, when developing BNB (otherwise leave that out)</span><!-- HTML_TAG_END --></pre></div> <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-raeog1">If you have multiple versions of CUDA installed or installed it in a non-standard location, please refer to CMake CUDA documentation for how to configure the CUDA compiler.</p></div> </div> <h3 class="relative group"><a id="pytorch-cuda-versions" 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="#pytorch-cuda-versions"><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>PyTorch CUDA versions</span></h3> <p data-svelte-h="svelte-n0pdc4">Some bitsandbytes features may need a newer CUDA version than the one currently supported by PyTorch binaries from Conda and pip. In this case, you should follow these instructions to load a precompiled bitsandbytes binary.</p> <ol data-svelte-h="svelte-w9mxyz"><li>Determine the path of the CUDA version you want to use. Common paths include:</li></ol> <ul data-svelte-h="svelte-r36v8e"><li><code>/usr/local/cuda</code></li> <li><code>/usr/local/cuda-XX.X</code> where <code>XX.X</code> is the CUDA version number</li></ul> <p data-svelte-h="svelte-1dhvzk8">Then locally install the CUDA version you need with this script from bitsandbytes:</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 -->wget https://raw.githubusercontent.com/bitsandbytes-foundation/bitsandbytes/main/install_cuda.sh | |
| <span class="hljs-comment"># Syntax cuda_install CUDA_VERSION INSTALL_PREFIX EXPORT_TO_BASH</span> | |
| <span class="hljs-comment"># CUDA_VERSION in {110, 111, 112, 113, 114, 115, 116, 117, 118, 120, 121, 122, 123, 124, 125}</span> | |
| <span class="hljs-comment"># EXPORT_TO_BASH in {0, 1} with 0=False and 1=True</span> | |
| <span class="hljs-comment"># For example, the following installs CUDA 11.7 to ~/local/cuda-11.7 and exports the path to your .bashrc</span> | |
| bash install_cuda.sh 117 ~/local 1<!-- HTML_TAG_END --></pre></div> <ol start="2" data-svelte-h="svelte-y8zpl2"><li>Set the environment variables <code>BNB_CUDA_VERSION</code> and <code>LD_LIBRARY_PATH</code> by manually overriding the CUDA version installed by PyTorch.</li></ol> <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-143buxs">It is recommended to add the following lines to the <code>.bashrc</code> file to make them permanent.</p></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 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-built_in">export</span> BNB_CUDA_VERSION=<VERSION> | |
| <span class="hljs-built_in">export</span> LD_LIBRARY_PATH=<span class="hljs-variable">$LD_LIBRARY_PATH</span>:<PATH><!-- HTML_TAG_END --></pre></div> <p data-svelte-h="svelte-wx07f3">For example, to use a local install path:</p> <div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-built_in">export</span> BNB_CUDA_VERSION=117 | |
| <span class="hljs-built_in">export</span> LD_LIBRARY_PATH=<span class="hljs-variable">$LD_LIBRARY_PATH</span>:/home/YOUR_USERNAME/local/cuda-11.7<!-- HTML_TAG_END --></pre></div> <ol start="3" data-svelte-h="svelte-lfqu8"><li>Now when you launch bitsandbytes with these environment variables, the PyTorch CUDA version is overridden by the new CUDA version (in this example, version 11.7) and a different bitsandbytes library is loaded.</li></ol> <h2 class="relative group"><a id="multi-backend" 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="#multi-backend"><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>Multi-backend Support (Alpha Release)</span></h2> <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-178d4s9">This functionality is currently in preview and not yet production-ready. We very much welcome community feedback, contributions and leadership on topics like Apple Silicon as well as other less common accellerators! For more information, see <a href="./non_cuda_backends">this guide on multi-backend support</a>.</p></div> <p data-svelte-h="svelte-1jdzqrl"><strong>Link to give us feedback</strong> (bugs, install issues, perf results, requests, etc.)<strong>:</strong></p> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">ROCm </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Intel CPU+GPU </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Apple Silicon / Metal (MPS) </div></div> <div class="language-select"><p data-svelte-h="svelte-1jd3851"><a href="https://github.com/bitsandbytes-foundation/bitsandbytes/discussions/1339" rel="nofollow"><strong>Multi-backend refactor: Alpha release (AMD ROCm ONLY)</strong></a></p> </div> <h3 class="relative group"><a id="multi-backend-supported-backends" 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="#multi-backend-supported-backends"><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>Supported Backends</span></h3> <table data-svelte-h="svelte-yxyu4i"><thead><tr><th><strong>Backend</strong></th> <th><strong>Supported Versions</strong></th> <th><strong>Python versions</strong></th> <th><strong>Architecture Support</strong></th> <th><strong>Status</strong></th></tr></thead> <tbody><tr><td><strong>AMD ROCm</strong></td> <td>6.1+</td> <td>3.10+</td> <td>minimum CDNA - <code>gfx90a</code>, RDNA - <code>gfx1100</code></td> <td>Alpha</td></tr> <tr><td><strong>Apple Silicon (MPS)</strong></td> <td>WIP</td> <td>3.10+</td> <td>M1/M2 chips</td> <td>Planned</td></tr> <tr><td><strong>Intel CPU</strong></td> <td>v2.4.0+ (<code>ipex</code>)</td> <td>3.10+</td> <td>Intel CPU</td> <td>Alpha</td></tr> <tr><td><strong>Intel GPU</strong></td> <td>v2.4.0+ (<code>ipex</code>)</td> <td>3.10+</td> <td>Intel GPU</td> <td>Experimental</td></tr></tbody></table> <p data-svelte-h="svelte-19x599i">For each supported backend, follow the respective instructions below:</p> <h3 class="relative group"><a id="multi-backend-pre-requisites" 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="#multi-backend-pre-requisites"><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>Pre-requisites</span></h3> <p data-svelte-h="svelte-abm2bq">To use bitsandbytes non-CUDA backends, be sure to install:</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 <span class="hljs-keyword">install</span> <span class="hljs-string">"transformers>=4.45.1"</span><!-- HTML_TAG_END --></pre></div> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">AMD ROCm </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Intel CPU + GPU </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Apple Silicon (MPS) </div></div> <div class="language-select"> <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-qw1mn7">Pre-compiled binaries are only built for ROCm versions <code>6.1.0</code>/<code>6.1.1</code>/<code>6.1.2</code>/<code>6.2.0</code> and <code>gfx90a</code>, <code>gfx942</code>, <code>gfx1100</code> GPU architectures. <a href="#multi-backend-pip">Find the pip install instructions here</a>.</p> <p data-svelte-h="svelte-1yin7gn">Other supported versions that don’t come with pre-compiled binaries <a href="#multi-backend-compile">can be compiled for with these instructions</a>.</p> <p data-svelte-h="svelte-5vnmxq"><strong>Windows is not supported for the ROCm backend</strong>; also not WSL2 to our knowledge.</p></div> <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-1ioez28">If you would like to install ROCm and PyTorch on bare metal, skip the Docker steps and refer to ROCm’s official guides at <a href="https://rocm.docs.amd.com/projects/install-on-linux/en/latest/tutorial/install-overview.html#rocm-install-overview" rel="nofollow">ROCm installation overview</a> and <a href="https://rocm.docs.amd.com/projects/install-on-linux/en/latest/how-to/3rd-party/pytorch-install.html#using-wheels-package" rel="nofollow">Installing PyTorch for ROCm</a> (Step 3 of wheels build for quick installation). Special note: please make sure to get the respective ROCm-specific PyTorch wheel for the installed ROCm version, e.g. <code>https://download.pytorch.org/whl/nightly/rocm6.2/</code>!</p></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 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-comment"># Create a docker container with latest ROCm image, which includes ROCm libraries</span> | |
| docker pull rocm/dev-ubuntu-22.04:6.1.2-complete | |
| docker run -it --device=/dev/kfd --device=/dev/dri --group-add video rocm/dev-ubuntu-22.04:6.1.2-complete | |
| apt-get update && apt-get install -y git && <span class="hljs-built_in">cd</span> home | |
| <span class="hljs-comment"># Install pytorch compatible with above ROCm version</span> | |
| pip install torch --index-url https://download.pytorch.org/whl/rocm6.1/<!-- HTML_TAG_END --></pre></div> </div> <h3 class="relative group"><a id="installation" 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="#installation"><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>Installation</span></h3> <p data-svelte-h="svelte-pvlxk7">You can install the pre-built wheels for each backend, or compile from source for custom configurations.</p> <h4 class="relative group"><a id="multi-backend-pip" 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="#multi-backend-pip"><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>Pre-built Wheel Installation (recommended)</span></h4> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">Linux </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Windows </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Mac </div></div> <div class="language-select"><div class="code-block relative"><div class="absolute top-2.5 right-4"><button class="inline-flex items-center relative text-sm focus:text-green-500 cursor-pointer focus:outline-none transition duration-200 ease-in-out opacity-0 mx-0.5 text-gray-600 " title="code excerpt" type="button"><svg class="" xmlns="http://www.w3.org/2000/svg" aria-hidden="true" fill="currentColor" focusable="false" role="img" width="1em" height="1em" preserveAspectRatio="xMidYMid meet" viewBox="0 0 32 32"><path d="M28,10V28H10V10H28m0-2H10a2,2,0,0,0-2,2V28a2,2,0,0,0,2,2H28a2,2,0,0,0,2-2V10a2,2,0,0,0-2-2Z" transform="translate(0)"></path><path d="M4,18H2V4A2,2,0,0,1,4,2H18V4H4Z" transform="translate(0)"></path><rect fill="none" width="32" height="32"></rect></svg> <div class="absolute pointer-events-none transition-opacity bg-black text-white py-1 px-2 leading-tight rounded font-normal shadow left-1/2 top-full transform -translate-x-1/2 translate-y-2 opacity-0"><div class="absolute bottom-full left-1/2 transform -translate-x-1/2 w-0 h-0 border-black border-4 border-t-0" style="border-left-color: transparent; border-right-color: transparent; "></div> Copied</div></button></div> <pre class=""><!-- HTML_TAG_START --><span class="hljs-comment"># Note, if you don't want to reinstall BNBs dependencies, append the `--no-deps` flag!</span> | |
| <span class="hljs-attribute">pip</span> install --force-reinstall 'https://github.com/bitsandbytes-foundation/bitsandbytes/releases/download/continuous-release_multi-backend-refactor/bitsandbytes-<span class="hljs-number">0</span>.<span class="hljs-number">44</span>.<span class="hljs-number">1</span>.dev0-py3-none-manylinux_2_24_x86_64.whl'<!-- HTML_TAG_END --></pre></div> </div> <h4 class="relative group"><a id="multi-backend-compile" 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="#multi-backend-compile"><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>Compile from Source</span></h4> <div class="flex space-x-2 items-center my-1.5 mr-8 h-7 !pl-0 -mx-3 md:mx-0"><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd border-gray-800 bg-black dark:bg-gray-700 text-white">AMD ROCm </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Intel CPU + GPU </div><div class="flex items-center border rounded-lg px-1.5 py-1 leading-none select-none text-smd text-gray-500 cursor-pointer opacity-90 hover:text-gray-700 dark:hover:text-gray-200 hover:shadow-sm">Apple Silicon (MPS) </div></div> <div class="language-select"> <h4 class="relative group"><a id="amd-gpu" 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="#amd-gpu"><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>AMD GPU</span></h4> <p data-svelte-h="svelte-1fw7vjl">bitsandbytes is fully supported from ROCm 6.1 onwards (currently in alpha release).</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-comment"># Install bitsandbytes from source</span> | |
| <span class="hljs-comment"># Clone bitsandbytes repo, ROCm backend is currently enabled on multi-backend-refactor branch</span> | |
| git <span class="hljs-built_in">clone</span> -b multi-backend-refactor https://github.com/bitsandbytes-foundation/bitsandbytes.git && <span class="hljs-built_in">cd</span> bitsandbytes/ | |
| <span class="hljs-comment"># Install dependencies</span> | |
| pip install -r requirements-dev.txt | |
| <span class="hljs-comment"># Compile & install</span> | |
| apt-get install -y build-essential cmake <span class="hljs-comment"># install build tools dependencies, unless present</span> | |
| cmake -DCOMPUTE_BACKEND=hip -S . <span class="hljs-comment"># Use -DBNB_ROCM_ARCH="gfx90a;gfx942" to target specific gpu arch</span> | |
| make | |
| pip install -e . <span class="hljs-comment"># `-e` for "editable" install, when developing BNB (otherwise leave that out)</span><!-- HTML_TAG_END --></pre></div> </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/installation.mdx" target="_blank"><span data-svelte-h="svelte-1kd6by1"><</span> <span data-svelte-h="svelte-x0xyl0">></span> <span data-svelte-h="svelte-1dajgef"><span class="underline ml-1.5">Update</span> on GitHub</span></a> <p></p> | |
| <script> | |
| { | |
| __sveltekit_1amjjgw = { | |
| assets: "/docs/bitsandbytes/pr_1137/en", | |
| base: "/docs/bitsandbytes/pr_1137/en", | |
| env: {} | |
| }; | |
| const element = document.currentScript.parentElement; | |
| const data = [null,null]; | |
| Promise.all([ | |
| import("/docs/bitsandbytes/pr_1137/en/_app/immutable/entry/start.435ebb3a.js"), | |
| import("/docs/bitsandbytes/pr_1137/en/_app/immutable/entry/app.6b2aadb8.js") | |
| ]).then(([kit, app]) => { | |
| kit.start(app, element, { | |
| node_ids: [0, 10], | |
| data, | |
| form: null, | |
| error: null | |
| }); | |
| }); | |
| } | |
| </script> | |
Xet Storage Details
- Size:
- 55.8 kB
- Xet hash:
- 4216f1732d409086a4521c4ceb43c4b422272acbf2b2e29b4b05e0bd3b68d6cc
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.