Files
doc-exports/docs/modelarts/umn/develop-modelarts-0079.html
Lai, Weijian 6aa966a79a ModelArts UMN 24.3.0 version
Reviewed-by: Pruthi, Vineet <vineet.pruthi@t-systems.com>
Co-authored-by: Lai, Weijian <laiweijian4@huawei.com>
Co-committed-by: Lai, Weijian <laiweijian4@huawei.com>
2024-11-02 09:04:52 +00:00

17 lines
3.0 KiB
HTML

<a name="EN-US_TOPIC_0000002079181641"></a><a name="EN-US_TOPIC_0000002079181641"></a>
<h1 class="topictitle1">Specifications for Custom Images Used for Training Jobs</h1>
<div id="body8662426"><p id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_p1037492192113">When you use a locally developed model or training script to create a custom image, ensure that the custom image complies with the specifications defined by ModelArts.</p>
<div class="section" id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_section26795139453"><h4 class="sectiontitle">Specifications</h4><ul id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_ul557515200535"><li class="SourceCode" id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_li18576520195314">A custom image cannot be larger than 30 GB. It is best if the custom image is no larger than 15 GB. An oversized image affects the startup of a training job.</li></ul>
<ul id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_ul142648262454"><li id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_li88571413163219">The <span class="parmname" id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_parmname1450252503018"><b>uid</b></span> of the default user of a custom image must be <span class="parmvalue" id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_parmvalue166371629153011"><b>1000</b></span>.</li><li id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_li17932924101616">GPU drivers cannot be installed in an custom image. When you use GPUs to run a training job, ModelArts automatically places the GPU driver in <strong id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_b193534166413">/usr/local/nvidia</strong> of the training environment.</li><li id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_li46120446715">ModelArts does not support the download of open source installation packages. Install the dependencies required by the training job in the custom image.</li></ul>
</div>
<div class="section" id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_section1712505175116"><h4 class="sectiontitle">Creating a Custom Image for Training Jobs</h4><ol id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_en-us_topic_0171858297_ol108388303371"><li id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_en-us_topic_0171858297_li6838173019379">Set up the Docker environment on an ECS or a local host.</li><li id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_en-us_topic_0171858297_li283814309373">Write a Dockerfile based on your needs and the preceding specifications to build a custom image.</li><li id="EN-US_TOPIC_0000002079181641__en-us_topic_0000001799498568_li16878054121612">Upload the created custom image to SWR.</li></ol>
</div>
</div>
<div>
<div class="familylinks">
<div class="parentlink"><strong>Parent topic:</strong> <a href="docker-modelarts_0017.html">Using a Custom Image to Train Models (New-Version Training)</a></div>
</div>
</div>