Deployed 6189ef2 to 2.0 with MkDocs 1.6.0 and mike 2.1.3

Author: gitlawr

2.0/installation/amd/installation/index.html

@@ -3860,6 +3860,7 @@ <h2 id="installation">Installation</h2>
 <div class="highlight"><pre><span></span><code>sudo<span class="w"> </span>docker<span class="w"> </span>run<span class="w"> </span>-d<span class="w"> </span>--name<span class="w"> </span>gpustack<span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--restart<span class="w"> </span>unless-stopped<span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--privileged<span class="w"> </span><span class="se">\</span>
+<span class="w">    </span>--volume<span class="w"> </span>/opt/rocm:/opt/rocm:ro<span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--network<span class="w"> </span>host<span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--volume<span class="w"> </span>/var/run/docker.sock:/var/run/docker.sock<span class="w"> </span><span class="se">\</span>
 <span class="w">    </span>--volume<span class="w"> </span>gpustack-data:/var/lib/gpustack<span class="w"> </span><span class="se">\</span>
@@ -3870,6 +3871,7 @@ <h2 id="installation">Installation</h2>
 <li>The <code>--privileged</code> flag is required for device vendor-agnostic access.</li>
 <li>To restrict GPU access, remove <code>--privileged</code> flag and set the <code>AMD_VISIBLE_DEVICES</code> environment variable.
   See <a href="https://instinct.docs.amd.com/projects/container-toolkit/en/latest/container-runtime/migration-guide.html">AMD Container Runtime - Migration Guide</a>.</li>
+<li>If the <code>/opt/rocm</code> directory does not exist, please create a symbolic link pointing to the ROCm installed path: <code>ln -s /path/to/rocm /opt/rocm</code>.</li>
 <li>The <code>--network=host</code> option is necessary for port awareness.</li>
 <li>Mounting <code>/var/run/docker.sock</code> allows GPUStack to manage Docker containers for inference engines.</li>
 </ul>

2.0/installation/ascend/installation/index.html

@@ -3873,7 +3873,20 @@ <h2 id="installation">Installation</h2>
 </code></pre></div>
 <ul>
 <li>To restrict NPU access, remove <code>--privileged</code> flag and set the <code>ASCEND_VISIBLE_DEVICES</code> environment variable.
-  See <a href="https://www.hiascend.com/document/detail/zh/mindcluster/72rc1/clustersched/dlug/dlruntime_ug_004.html">MindCluster - Docker Client Usage</a>.</li>
+  See <a href="https://www.hiascend.com/document/detail/zh/mindcluster/72rc1/clustersched/dlug/dlruntime_ug_004.html">MindCluster - Docker Client Usage</a>.
+  When GPUStack detects device indexes that don't start from 0 or are not consecutive, it performs automatic alignment. 
+  This is useful for virtual machine scenarios where only some devices are pass-through. 
+  However, if the host's devices are consecutive, automatic alignment can be counterproductive. 
+  Therefore, automatic alignment can be turned off to avoid wrong device mapping.</li>
+</ul>
+<div class="highlight"><pre><span></span><code><span class="w"> </span>sudo docker run -d --name gpustack \
+<span class="w"> </span>    ...
+<span class="gd">-    --privileged \</span>
+<span class="gi">+    --env &quot;ASCEND_VISIBLE_DEVICES=&lt;expected_devices_comma_seperated_list&gt;&quot; \</span>
+<span class="gi">+    --env &quot;GPUSTACK_RUNTIME_DEPLOY_BACKEND_VISIBLE_DEVICES_VALUE_ALIGNMENT=&quot; \</span>
+<span class="w"> </span>    ...
+</code></pre></div>
+<ul>
 <li>The <code>--network=host</code> option is necessary for port awareness.</li>
 <li>Mounting <code>/var/run/docker.sock</code> allows GPUStack to manage Docker containers for inference engines.</li>
 </ul>