https://iwmei.top/categories/pytorch/Recent content in PyTorch on iwmei的博客Hugo -- gohugo.iozh-cniwmeiSun, 08 Mar 2026 00:00:00 +0000https://iwmei.top/p/windows-torch-management/Sun, 08 Mar 2026 00:00:00 +0000https://iwmei.top/p/windows-torch-management/<img src="https://iwmei.top/p/windows-torch-management/cover.png" alt="Featured image of post Windows 上有关 Torch 的版本管理问题" /><h1 id="windows上有关torch的版本管理问题">Windows上有关Torch的版本管理问题 </h1><h2 id="不要完全相信ai的话">不要完全相信AI的话 </h2><ul> <li>AI在此领域的资料有<strong>明显滞后性</strong>；</li> <li>AI处理环境问题显得非常力不从心，容易陷入<strong>思维定势</strong>；</li> <li>一旦AI然你<strong>重新配置环境</strong>，立刻放弃继续询问AI。</li> </ul> <h2 id="配置环境的正确步骤">配置环境的正确步骤 </h2><ol> <li> <p>如果你使用Nvidia显卡，请务必仔细；</p> </li> <li> <p>打开Pytorch官网链接：</p> <div class="highlight"><div class="chroma"> <table class="lntable"><tr><td class="lntd"> <pre tabindex="0" class="chroma"><code><span class="lnt">1 </span></code></pre></td> <td class="lntd"> <pre tabindex="0" class="chroma"><code class="language-link" data-lang="link"><span class="line"><span class="cl"><span class="err">[https://pytorch.org/get-started/locally/](https://pytorch.org/get-started/locally/)</span> </span></span></code></pre></td></tr></table> </div> </div></li> <li> <p>在页面中选择与自身机器匹配的版本：</p> <div class="highlight"><div class="chroma"> <table class="lntable"><tr><td class="lntd"> <pre tabindex="0" class="chroma"><code><span class="lnt">1 </span></code></pre></td> <td class="lntd"> <pre tabindex="0" class="chroma"><code class="language-link" data-lang="link"><span class="line"><span class="cl"><span class="err">nvidia-smi</span> <span class="c"># 在终端执行此命令，查看CUDA版本，若为非Nvidia显卡，则忽略此条</span> </span></span></code></pre></td></tr></table> </div> </div><div class="highlight"><div class="chroma"> <table class="lntable"><tr><td class="lntd"> <pre tabindex="0" class="chroma"><code><span class="lnt">1 </span></code></pre></td> <td class="lntd"> <pre tabindex="0" class="chroma"><code class="language-link" data-lang="link"><span class="line"><span class="cl"><span class="err">pip3</span> <span class="err">install</span> <span class="err">torch</span> <span class="err">torchaudio</span> <span class="err">torchvision</span> <span class="err">--index-url</span> <span class="err">[https://download.pytorch.org/whl/cu130](https://download.pytorch.org/whl/cu130)</span> <span class="c"># 根据我的电脑来的，pip3 可换成 pip 或 uv pip</span> </span></span></code></pre></td></tr></table> </div> </div></li> <li> <p>如果有<strong>torchcodec</strong>的需求：</p> <div class="highlight"><div class="chroma"> <table class="lntable"><tr><td class="lntd"> <pre tabindex="0" class="chroma"><code><span class="lnt">1 </span></code></pre></td> <td class="lntd"> <pre tabindex="0" class="chroma"><code class="language-link" data-lang="link"><span class="line"><span class="cl"><span class="na">[https://github.com/meta-pytorch/torchcodec?tab</span><span class="o">=</span><span class="s">readme-ov-file#installing-torchcodec](https://github.com/meta-pytorch/torchcodec?tab=readme-ov-file#installing-torchcodec)</span> </span></span></code></pre></td></tr></table> </div> </div><ul> <li>在这里查看与所安装的torch相匹配的版本。</li> </ul> </li> <li> <p>在环境目录下的终端中，运行安装命令。</p> <div class="highlight"><div class="chroma"> <table class="lntable"><tr><td class="lntd"> <pre tabindex="0" class="chroma"><code><span class="lnt">1 </span></code></pre></td> <td class="lntd"> <pre tabindex="0" class="chroma"><code class="language-link" data-lang="link"><span class="line"><span class="cl"><span class="err">uv</span> <span class="err">pip</span> <span class="err">install</span> <span class="err">torch</span> <span class="err">torchaudio</span> <span class="err">torchvision</span> <span class="err">--index-url</span> <span class="err">[https://download.pytorch.org/whl/cu130](https://download.pytorch.org/whl/cu130)</span> </span></span></code></pre></td></tr></table> </div> </div><div class="highlight"><div class="chroma"> <table class="lntable"><tr><td class="lntd"> <pre tabindex="0" class="chroma"><code><span class="lnt">1 </span></code></pre></td> <td class="lntd"> <pre tabindex="0" class="chroma"><code class="language-link" data-lang="link"><span class="line"><span class="cl"><span class="na">conda install -c conda-forge &#34;torchcodec</span><span class="o">=</span><span class="s">*=*cuda*&#34;</span> </span></span></code></pre></td></tr></table> </div> </div><ul> <li>注意，windows下的torchcodec需要在conda环境下安装，以支持调用cuda加速。</li> </ul> </li> </ol>