[{"data":1,"prerenderedAt":2290},["ShallowReactive",2],{"/blog/empowering-modelops-and-hpc-workloads-with-gpu-enabled-runners":3,"navigation-en-us":1539,"banner-en-us":1976,"footer-en-us":1984,"blog-post-authors-en-us-Gabriel Engel":2230,"blog-related-posts-en-us-empowering-modelops-and-hpc-workloads-with-gpu-enabled-runners":2245,"blog-share-wrapper-en-us":2270,"next-steps-en-us":2280},{"id":4,"title":5,"authors":6,"body":8,"category":1515,"date":1516,"description":1517,"extension":1518,"externalUrl":1519,"faq":1519,"featured":1520,"heroImage":1521,"hidePromo":1520,"meta":1522,"navigation":207,"path":1523,"seo":1524,"slug":1529,"stem":1530,"tags":1531,"template":1537,"updatedDate":1519,"__hash__":1538},"blogPosts/en-us/blog/empowering-modelops-and-hpc-workloads-with-gpu-enabled-runners.md","Empower ModelOps and HPC workloads with GPU-enabled runners integrated with CI/CD",[7],"Gabriel Engel",{"type":9,"value":10,"toc":1497},"minimark",[11,29,44,47,52,72,76,79,84,101,105,125,177,181,195,198,219,223,234,257,261,264,268,275,280,371,375,378,382,435,439,442,446,455,474,477,480,520,527,532,550,658,675,823,832,857,873,879,883,893,899,967,978,1035,1039,1042,1048,1133,1138,1144,1150,1154,1160,1225,1237,1243,1247,1256,1274,1279,1440,1446,1454,1463,1467,1470,1478,1493],[12,13,14],"p",{},[15,16,17,18,23,24,28],"em",{},"This blog post is the latest in an ongoing series about GitLab's journey to ",[19,20,22],"a",{"href":21},"/blog/ai-ml-in-devsecops-series/","build and integrate AI/ML into our DevSecOps platform",". Start with the first blog post: ",[19,25,27],{"href":26},"/blog/what-the-ml-ai/","What the ML is up with DevSecOps and AI?",". Throughout the series, we'll feature blogs from our product, engineering, and UX teams to showcase how we're infusing AI/ML into GitLab.",[12,30,31,32,37,38,43],{},"In today's fast-paced world, organizations are constantly looking to improve their ",[19,33,36],{"href":34,"rel":35},"https://docs.gitlab.com/user/project/ml/",[],"ModelOps"," and high-performance computing (HPC) capabilities. Leveraging powerful graphical processing units (",[19,39,42],{"href":40,"rel":41},"https://www.techtarget.com/searchvirtualdesktop/definition/GPU-graphics-processing-unit",[],"GPUs",") has become a game-changer for accelerating machine learning workflows and compute-intensive tasks. To help meet these evolving needs, we recently released our first GPU-enabled runners on GitLab.com.",[12,45,46],{},"Securely hosting a GitLab Runner environment for ModelOps and HPC is non-trivial and requires a lot of knowledge and time to set up and maintain. In this blog post, we'll look at some real-world examples of how you can harness the potential of GPU computing for ModelOps or HPC workloads while taking full advantage of a SaaS solution.",[48,49,51],"h2",{"id":50},"what-are-gpu-enabled-runners","What are GPU-enabled runners?",[12,53,54,55,60,61,65,66,71],{},"GPU-enabled runners are dedicated computing resources for the AI-powered DevSecOps platform. They provide accelerated processing power for ModelOps and HPC such as the training or deployment of large language models (",[19,56,59],{"href":57,"rel":58},"https://www.techtarget.com/whatis/definition/large-language-model-LLM",[],"LLMs",") as part of ModelOps workloads. In the first iteration of releasing GPU-enabled runners, GitLab.com SaaS offers the GCP ",[62,63,64],"code",{},"n1-standard-4"," machine type (4 vCPU, 15 GB memory) with 1 NVIDIA T4 (16 GB memory) attached. The runner behaves like a GitLab Runner on Linux, using the docker+machine ",[19,67,70],{"href":68,"rel":69},"https://docs.gitlab.com/runner/executors/",[],"executor",".",[48,73,75],{"id":74},"using-gpu-enabled-runners","Using GPU-enabled runners",[12,77,78],{},"To take advantage of GitLab GPU-enabled runners, follow these steps:",[80,81,83],"h3",{"id":82},"_1-have-a-project-on-gitlabcom","1. Have a project on GitLab.com",[12,85,86,87,90,91,94,95,100],{},"All projects on GitLab.com SaaS with a ",[62,88,89],{},"Premium"," or ",[62,92,93],{},"Ultimate"," ",[19,96,99],{"href":97,"rel":98},"https://about.gitlab.com/pricing/",[],"subscription"," have the GPU-enabled runners enabled by default - no additional configuration is required.",[80,102,104],{"id":103},"_2-create-a-job-running-on-gpu-enabled-runners","2. Create a job running on GPU-enabled runners",[12,106,107,108,111,112,120,121,124],{},"Create a job in your ",[62,109,110],{},".gitlab-ci.yml"," configuration file, and set the ",[19,113,116,117],{"href":114,"rel":115},"https://docs.gitlab.com/ci/runners/configure_runners/#use-tags-to-control-which-jobs-a-runner-can-run",[],"runner ",[62,118,119],{},"tag"," to the ",[62,122,123],{},"saas-linux-medium-amd64-gpu-standard"," value.",[126,127,132],"pre",{"className":128,"code":129,"language":130,"meta":131,"style":131},"language-yaml shiki shiki-themes github-light","gpu-job:\n  stage: build\n  tags:\n    - saas-linux-medium-amd64-gpu-standard\n\n","yaml","",[62,133,134,147,160,168],{"__ignoreMap":131},[135,136,139,143],"span",{"class":137,"line":138},"line",1,[135,140,142],{"class":141},"shJU0","gpu-job",[135,144,146],{"class":145},"sgsFI",":\n",[135,148,150,153,156],{"class":137,"line":149},2,[135,151,152],{"class":141},"  stage",[135,154,155],{"class":145},": ",[135,157,159],{"class":158},"sYBdl","build\n",[135,161,163,166],{"class":137,"line":162},3,[135,164,165],{"class":141},"  tags",[135,167,146],{"class":145},[135,169,171,174],{"class":137,"line":170},4,[135,172,173],{"class":145},"    - ",[135,175,176],{"class":158},"saas-linux-medium-amd64-gpu-standard\n",[80,178,180],{"id":179},"_3-select-a-docker-image-with-the-nvidia-cuda-driver","3. Select a Docker image with the Nvidia CUDA driver",[12,182,183,184,189,190,71],{},"The CI/CD job runs in an isolated virtual machine (VM) with a bring-your-own-image policy as with GitLab SaaS runners on Linux. GitLab mounts the GPU from the host VM into your isolated environment. You must use a Docker image with the GPU driver installed to use the GPU. For Nvidia GPUs, you can use the ",[19,185,188],{"href":186,"rel":187},"https://catalog.ngc.nvidia.com/orgs/nvidia/containers/cuda",[],"CUDA Toolkit"," directly, or third-party images with Nvidia drivers installed, such as the ",[19,191,194],{"href":192,"rel":193},"https://hub.docker.com/r/tensorflow/tensorflow/",[],"TensorFlow GPU image",[12,196,197],{},"The CI/CD job configuration for the Nvidia CUDA base Ubuntu image looks like this:",[126,199,201],{"className":128,"code":200,"language":130,"meta":131,"style":131},"\n  image: nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04\n\n",[62,202,203,209],{"__ignoreMap":131},[135,204,205],{"class":137,"line":138},[135,206,208],{"emptyLinePlaceholder":207},true,"\n",[135,210,211,214,216],{"class":137,"line":149},[135,212,213],{"class":141},"  image",[135,215,155],{"class":145},[135,217,218],{"class":158},"nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04\n",[80,220,222],{"id":221},"_4-verify-that-the-gpu-is-working","4. Verify that the GPU is working",[12,224,225,226,229,230,233],{},"To verify that the GPU drivers are working correctly, you can execute the ",[62,227,228],{},"nvidia-smi"," command in the CI/CD job ",[62,231,232],{},"script"," section.",[126,235,237],{"className":128,"code":236,"language":130,"meta":131,"style":131},"\n  script:\n    - nvidia-smi\n\n",[62,238,239,243,250],{"__ignoreMap":131},[135,240,241],{"class":137,"line":138},[135,242,208],{"emptyLinePlaceholder":207},[135,244,245,248],{"class":137,"line":149},[135,246,247],{"class":141},"  script",[135,249,146],{"class":145},[135,251,252,254],{"class":137,"line":162},[135,253,173],{"class":145},[135,255,256],{"class":158},"nvidia-smi\n",[48,258,260],{"id":259},"basic-usage-examples","Basic usage examples",[12,262,263],{},"Let's explore some basic scenarios where GPU-enabled runners can supercharge your ModelOps and HPC workloads:",[80,265,267],{"id":266},"example-1-modelops-with-python","Example 1: ModelOps with Python",[12,269,270,271,274],{},"In this example, we train a model on our GPU-enabled runner defined in the ",[62,272,273],{},"train.py"," file using the Nvidia CUDA base Ubuntu image mentioned earlier.",[12,276,277,279],{},[62,278,110],{}," file:",[126,281,283],{"className":128,"code":282,"language":130,"meta":131,"style":131},"model-training:\n  stage: build\n  tags:\n    - saas-linux-medium-amd64-gpu-standard\n  image: nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04\n  script:\n    - apt update\n    - apt install -y --no-install-recommends python3 python3-pip \n    - pip3 install -r requirements.txt\n    - python3 --version\n    - python3 train.py\n\n",[62,284,285,292,300,306,312,321,328,336,347,355,363],{"__ignoreMap":131},[135,286,287,290],{"class":137,"line":138},[135,288,289],{"class":141},"model-training",[135,291,146],{"class":145},[135,293,294,296,298],{"class":137,"line":149},[135,295,152],{"class":141},[135,297,155],{"class":145},[135,299,159],{"class":158},[135,301,302,304],{"class":137,"line":162},[135,303,165],{"class":141},[135,305,146],{"class":145},[135,307,308,310],{"class":137,"line":170},[135,309,173],{"class":145},[135,311,176],{"class":158},[135,313,315,317,319],{"class":137,"line":314},5,[135,316,213],{"class":141},[135,318,155],{"class":145},[135,320,218],{"class":158},[135,322,324,326],{"class":137,"line":323},6,[135,325,247],{"class":141},[135,327,146],{"class":145},[135,329,331,333],{"class":137,"line":330},7,[135,332,173],{"class":145},[135,334,335],{"class":158},"apt update\n",[135,337,339,341,344],{"class":137,"line":338},8,[135,340,173],{"class":145},[135,342,343],{"class":158},"apt install -y --no-install-recommends python3 python3-pip",[135,345,346],{"class":145}," \n",[135,348,350,352],{"class":137,"line":349},9,[135,351,173],{"class":145},[135,353,354],{"class":158},"pip3 install -r requirements.txt\n",[135,356,358,360],{"class":137,"line":357},10,[135,359,173],{"class":145},[135,361,362],{"class":158},"python3 --version\n",[135,364,366,368],{"class":137,"line":365},11,[135,367,173],{"class":145},[135,369,370],{"class":158},"python3 train.py\n",[80,372,374],{"id":373},"example-2-scientific-simulations-and-hpc","Example 2: Scientific simulations and HPC",[12,376,377],{},"Complex scientific simulations require significant computing resources. GPU-enabled runners can accelerate these simulations, allowing you to get results in less time.",[12,379,380,279],{},[62,381,110],{},[126,383,385],{"className":128,"code":384,"language":130,"meta":131,"style":131},"simulation-run:\n  stage: build\n  tags:\n    - saas-linux-medium-amd64-gpu-standard\n  image: nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04\n  script:\n    - ./run_simulation --input input_file.txt\n\n",[62,386,387,394,402,408,414,422,428],{"__ignoreMap":131},[135,388,389,392],{"class":137,"line":138},[135,390,391],{"class":141},"simulation-run",[135,393,146],{"class":145},[135,395,396,398,400],{"class":137,"line":149},[135,397,152],{"class":141},[135,399,155],{"class":145},[135,401,159],{"class":158},[135,403,404,406],{"class":137,"line":162},[135,405,165],{"class":141},[135,407,146],{"class":145},[135,409,410,412],{"class":137,"line":170},[135,411,173],{"class":145},[135,413,176],{"class":158},[135,415,416,418,420],{"class":137,"line":314},[135,417,213],{"class":141},[135,419,155],{"class":145},[135,421,218],{"class":158},[135,423,424,426],{"class":137,"line":323},[135,425,247],{"class":141},[135,427,146],{"class":145},[135,429,430,432],{"class":137,"line":330},[135,431,173],{"class":145},[135,433,434],{"class":158},"./run_simulation --input input_file.txt\n",[48,436,438],{"id":437},"advanced-usage-examples","Advanced usage examples",[12,440,441],{},"Let's go through some real-world scenarios of how we use GPU-enabled runners at GitLab.",[80,443,445],{"id":444},"example-3-python-model-training-with-a-custom-docker-image","Example 3: Python model training with a custom Docker image",[12,447,448,449,454],{},"For our third example, we will use this ",[19,450,453],{"href":451,"rel":452},"https://gitlab.com/gitlab-org/modelops/demos/handwritten-digit-recognition",[],"handwritten digit recognition model",". We are using this project as a demo to showcase or try out new ModelOps features.",[12,456,457,461,462,467,468,471,472,71],{},[19,458,460],{"href":451,"rel":459},[],"Open the project"," and fork it into your preferred namespace. You can follow the next steps using the ",[19,463,466],{"href":464,"rel":465},"https://docs.gitlab.com/user/project/web_ide/",[],"Web IDE"," in the browser, or clone the project locally to create and edit the files. Some of the next steps require you to override existing configuration in the ",[62,469,470],{},"Dockerfile"," and ",[62,473,110],{},[12,475,476],{},"As we need more pre-installed components and want to save installation time when training the model, we decided to create a custom Docker image with all dependencies pre-installed. This also gives us full control over the build environment we use and allows us to reuse it locally without relying on the `.gitlab-ci.yml' implementation.",[12,478,479],{},"In addition, we are using a more complete pipeline configuration with the following stages:",[126,481,483],{"className":128,"code":482,"language":130,"meta":131,"style":131},"stages:\n  - build\n  - test\n  - train\n  - publish\n\n",[62,484,485,492,499,506,513],{"__ignoreMap":131},[135,486,487,490],{"class":137,"line":138},[135,488,489],{"class":141},"stages",[135,491,146],{"class":145},[135,493,494,497],{"class":137,"line":149},[135,495,496],{"class":145},"  - ",[135,498,159],{"class":158},[135,500,501,503],{"class":137,"line":162},[135,502,496],{"class":145},[135,504,505],{"class":158},"test\n",[135,507,508,510],{"class":137,"line":170},[135,509,496],{"class":145},[135,511,512],{"class":158},"train\n",[135,514,515,517],{"class":137,"line":314},[135,516,496],{"class":145},[135,518,519],{"class":158},"publish\n",[12,521,522],{},[523,524],"img",{"alt":525,"src":526},"GPU pipeline overview","https://res.cloudinary.com/about-gitlab-com/image/upload/v1782397795/blog/Content%20Images/2023-07-06-gpu-enabled-runners-for-modelops/pipeline-overview.png",[528,529,531],"h4",{"id":530},"building-a-custom-docker-image","Building a custom Docker image",[12,533,534,535,537,538,541,542,545,546,549],{},"The first step is to define a ",[62,536,470],{},". In this example, we start with the Nvidia CUDA base Ubuntu image and then install ",[62,539,540],{},"Python3.10",". Using ",[62,543,544],{},"pip install",", we then add all the required libraries specified in a ",[62,547,548],{},"requirements.txt"," file.",[126,551,555],{"className":552,"code":553,"language":554,"meta":131,"style":131},"language-docker shiki shiki-themes github-light","FROM nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04\n\n1. Update and install required packages\nRUN apt-get update && apt-get install -y \\\n    python3.10 \\\n    python3.10-dev \\\n    python3-pip \\\n    && rm -rf /var/lib/apt/lists/*\n\n2. Set Python 3.10 as the default Python version\nRUN ln -s /usr/bin/python3.10 /usr/bin/python\n\n3. Copy the requirements.txt file\nCOPY requirements.txt /tmp/requirements.txt\n\n4. Install Python dependencies\nRUN pip3 install --no-cache-dir -r /tmp/requirements.txt\n","docker",[62,556,557,566,570,575,583,588,593,598,603,607,612,619,624,630,639,644,650],{"__ignoreMap":131},[135,558,559,563],{"class":137,"line":138},[135,560,562],{"class":561},"sD7c4","FROM",[135,564,565],{"class":145}," nvcr.io/nvidia/cuda:12.1.1-base-ubuntu22.04\n",[135,567,568],{"class":137,"line":149},[135,569,208],{"emptyLinePlaceholder":207},[135,571,572],{"class":137,"line":162},[135,573,574],{"class":145},"1. Update and install required packages\n",[135,576,577,580],{"class":137,"line":170},[135,578,579],{"class":561},"RUN",[135,581,582],{"class":145}," apt-get update && apt-get install -y \\\n",[135,584,585],{"class":137,"line":314},[135,586,587],{"class":145},"    python3.10 \\\n",[135,589,590],{"class":137,"line":323},[135,591,592],{"class":145},"    python3.10-dev \\\n",[135,594,595],{"class":137,"line":330},[135,596,597],{"class":145},"    python3-pip \\\n",[135,599,600],{"class":137,"line":338},[135,601,602],{"class":145},"    && rm -rf /var/lib/apt/lists/*\n",[135,604,605],{"class":137,"line":349},[135,606,208],{"emptyLinePlaceholder":207},[135,608,609],{"class":137,"line":357},[135,610,611],{"class":145},"2. Set Python 3.10 as the default Python version\n",[135,613,614,616],{"class":137,"line":365},[135,615,579],{"class":561},[135,617,618],{"class":145}," ln -s /usr/bin/python3.10 /usr/bin/python\n",[135,620,622],{"class":137,"line":621},12,[135,623,208],{"emptyLinePlaceholder":207},[135,625,627],{"class":137,"line":626},13,[135,628,629],{"class":145},"3. Copy the requirements.txt file\n",[135,631,633,636],{"class":137,"line":632},14,[135,634,635],{"class":561},"COPY",[135,637,638],{"class":145}," requirements.txt /tmp/requirements.txt\n",[135,640,642],{"class":137,"line":641},15,[135,643,208],{"emptyLinePlaceholder":207},[135,645,647],{"class":137,"line":646},16,[135,648,649],{"class":145},"4. Install Python dependencies\n",[135,651,653,655],{"class":137,"line":652},17,[135,654,579],{"class":561},[135,656,657],{"class":145}," pip3 install --no-cache-dir -r /tmp/requirements.txt\n",[12,659,660,661,663,664,669,670,71],{},"In the ",[62,662,110],{}," file we use ",[19,665,668],{"href":666,"rel":667},"https://docs.gitlab.com/ci/docker/using_kaniko/",[],"Kaniko"," to build the Docker image and push it into the ",[19,671,674],{"href":672,"rel":673},"https://docs.gitlab.com/user/packages/container_registry/",[],"GitLab Container Registry",[126,676,678],{"className":128,"code":677,"language":130,"meta":131,"style":131},"variables:\n  IMAGE_PATH: \"${CI_REGISTRY_IMAGE}:latest\"\n  GIT_STRATEGY: fetch\n\ndocker-build:\n  stage: build\n  tags:\n    - saas-linux-medium-amd64\n  image:\n    name: gcr.io/kaniko-project/executor:v1.9.0-debug\n    entrypoint: [\"\"]\n  script:\n    - /kaniko/executor\n      --context \"${CI_PROJECT_DIR}\"\n      --dockerfile \"${CI_PROJECT_DIR}/Dockerfile\"\n      --destination \"${IMAGE_PATH}\"\n      --destination \"${CI_REGISTRY_IMAGE}:${CI_COMMIT_TAG}\"\n  rules:\n    - if: $CI_COMMIT_TAG\n\n",[62,679,680,687,697,707,711,718,726,732,739,745,755,769,775,782,787,792,797,802,810],{"__ignoreMap":131},[135,681,682,685],{"class":137,"line":138},[135,683,684],{"class":141},"variables",[135,686,146],{"class":145},[135,688,689,692,694],{"class":137,"line":149},[135,690,691],{"class":141},"  IMAGE_PATH",[135,693,155],{"class":145},[135,695,696],{"class":158},"\"${CI_REGISTRY_IMAGE}:latest\"\n",[135,698,699,702,704],{"class":137,"line":162},[135,700,701],{"class":141},"  GIT_STRATEGY",[135,703,155],{"class":145},[135,705,706],{"class":158},"fetch\n",[135,708,709],{"class":137,"line":170},[135,710,208],{"emptyLinePlaceholder":207},[135,712,713,716],{"class":137,"line":314},[135,714,715],{"class":141},"docker-build",[135,717,146],{"class":145},[135,719,720,722,724],{"class":137,"line":323},[135,721,152],{"class":141},[135,723,155],{"class":145},[135,725,159],{"class":158},[135,727,728,730],{"class":137,"line":330},[135,729,165],{"class":141},[135,731,146],{"class":145},[135,733,734,736],{"class":137,"line":338},[135,735,173],{"class":145},[135,737,738],{"class":158},"saas-linux-medium-amd64\n",[135,740,741,743],{"class":137,"line":349},[135,742,213],{"class":141},[135,744,146],{"class":145},[135,746,747,750,752],{"class":137,"line":357},[135,748,749],{"class":141},"    name",[135,751,155],{"class":145},[135,753,754],{"class":158},"gcr.io/kaniko-project/executor:v1.9.0-debug\n",[135,756,757,760,763,766],{"class":137,"line":365},[135,758,759],{"class":141},"    entrypoint",[135,761,762],{"class":145},": [",[135,764,765],{"class":158},"\"\"",[135,767,768],{"class":145},"]\n",[135,770,771,773],{"class":137,"line":621},[135,772,247],{"class":141},[135,774,146],{"class":145},[135,776,777,779],{"class":137,"line":626},[135,778,173],{"class":145},[135,780,781],{"class":158},"/kaniko/executor\n",[135,783,784],{"class":137,"line":632},[135,785,786],{"class":158},"      --context \"${CI_PROJECT_DIR}\"\n",[135,788,789],{"class":137,"line":641},[135,790,791],{"class":158},"      --dockerfile \"${CI_PROJECT_DIR}/Dockerfile\"\n",[135,793,794],{"class":137,"line":646},[135,795,796],{"class":158},"      --destination \"${IMAGE_PATH}\"\n",[135,798,799],{"class":137,"line":652},[135,800,801],{"class":158},"      --destination \"${CI_REGISTRY_IMAGE}:${CI_COMMIT_TAG}\"\n",[135,803,805,808],{"class":137,"line":804},18,[135,806,807],{"class":141},"  rules",[135,809,146],{"class":145},[135,811,813,815,818,820],{"class":137,"line":812},19,[135,814,173],{"class":145},[135,816,817],{"class":141},"if",[135,819,155],{"class":145},[135,821,822],{"class":158},"$CI_COMMIT_TAG\n",[12,824,825,826,831],{},"In ",[19,827,830],{"href":828,"rel":829},"https://docs.gitlab.com/ci/yaml/#rules",[],"rules"," we define to only trigger the Docker image build for a new git tag. The reason is simple - we don't want to run the image build process for every time we train the model.",[12,833,834,835,840,841,844,845,848,849,852,853,856],{},"To start the image build job ",[19,836,839],{"href":837,"rel":838},"https://docs.gitlab.com/user/project/repository/tags/#create-a-tag",[],"create a new Git tag",". You can either do this by using ",[62,842,843],{},"git tag -a v0.0.1"," command or via UI. Navigate into ",[62,846,847],{},"Code > Tags"," and click on ",[62,850,851],{},"New Tag",". As Tag name type ",[62,854,855],{},"v0.0.1"," to create a new Git tag and trigger the job.",[12,858,859,860,863,864,866,867,71],{},"Navigate to ",[62,861,862],{},"Build > Pipelines"," to verify the ",[62,865,715],{}," job status, and then locate the tagged image following ",[19,868,870],{"href":672,"rel":869},[],[62,871,872],{},"Deploy > Container Registry",[12,874,875],{},[523,876],{"alt":877,"src":878},"Docker image","https://res.cloudinary.com/about-gitlab-com/image/upload/v1782397795/blog/Content%20Images/2023-07-06-gpu-enabled-runners-for-modelops/gpu-docker-image.png",[528,880,882],{"id":881},"testing-the-docker-image","Testing the Docker image",[12,884,885,886,889,890,892],{},"To test the image, we will use the following ",[62,887,888],{},"test-image"," job and run ",[62,891,228],{}," and check that the GPU drivers are working correctly.",[12,894,895,896,898],{},"The job configuration in ",[62,897,110],{}," file looks as follows:",[126,900,902],{"className":128,"code":901,"language":130,"meta":131,"style":131},"test-image:\n  stage: test\n  tags:\n    - saas-linux-medium-amd64-gpu-standard\n  image: $IMAGE_PATH\n  script:\n    - nvidia-smi\n  rules:\n    - if: $CI_COMMIT_TAG\n\n",[62,903,904,910,918,924,930,939,945,951,957],{"__ignoreMap":131},[135,905,906,908],{"class":137,"line":138},[135,907,888],{"class":141},[135,909,146],{"class":145},[135,911,912,914,916],{"class":137,"line":149},[135,913,152],{"class":141},[135,915,155],{"class":145},[135,917,505],{"class":158},[135,919,920,922],{"class":137,"line":162},[135,921,165],{"class":141},[135,923,146],{"class":145},[135,925,926,928],{"class":137,"line":170},[135,927,173],{"class":145},[135,929,176],{"class":158},[135,931,932,934,936],{"class":137,"line":314},[135,933,213],{"class":141},[135,935,155],{"class":145},[135,937,938],{"class":158},"$IMAGE_PATH\n",[135,940,941,943],{"class":137,"line":323},[135,942,247],{"class":141},[135,944,146],{"class":145},[135,946,947,949],{"class":137,"line":330},[135,948,173],{"class":145},[135,950,256],{"class":158},[135,952,953,955],{"class":137,"line":338},[135,954,807],{"class":141},[135,956,146],{"class":145},[135,958,959,961,963,965],{"class":137,"line":349},[135,960,173],{"class":145},[135,962,817],{"class":141},[135,964,155],{"class":145},[135,966,822],{"class":158},[12,968,969,970,975,976,549],{},"We also include container scanning and more ",[19,971,974],{"href":972,"rel":973},"https://docs.gitlab.com/user/application_security/",[],"security scanning"," templates in the ",[62,977,110],{},[126,979,981],{"className":128,"code":980,"language":130,"meta":131,"style":131},"include:\n  - template: Security/Secret-Detection.gitlab-ci.yml\n  - template: Security/Container-Scanning.gitlab-ci.yml\n  - template: Jobs/Dependency-Scanning.gitlab-ci.yml\n  - template: Security/SAST.gitlab-ci.yml\n\n",[62,982,983,990,1002,1013,1024],{"__ignoreMap":131},[135,984,985,988],{"class":137,"line":138},[135,986,987],{"class":141},"include",[135,989,146],{"class":145},[135,991,992,994,997,999],{"class":137,"line":149},[135,993,496],{"class":145},[135,995,996],{"class":141},"template",[135,998,155],{"class":145},[135,1000,1001],{"class":158},"Security/Secret-Detection.gitlab-ci.yml\n",[135,1003,1004,1006,1008,1010],{"class":137,"line":162},[135,1005,496],{"class":145},[135,1007,996],{"class":141},[135,1009,155],{"class":145},[135,1011,1012],{"class":158},"Security/Container-Scanning.gitlab-ci.yml\n",[135,1014,1015,1017,1019,1021],{"class":137,"line":170},[135,1016,496],{"class":145},[135,1018,996],{"class":141},[135,1020,155],{"class":145},[135,1022,1023],{"class":158},"Jobs/Dependency-Scanning.gitlab-ci.yml\n",[135,1025,1026,1028,1030,1032],{"class":137,"line":314},[135,1027,496],{"class":145},[135,1029,996],{"class":141},[135,1031,155],{"class":145},[135,1033,1034],{"class":158},"Security/SAST.gitlab-ci.yml\n",[528,1036,1038],{"id":1037},"training-the-model-with-our-custom-docker-image","Training the model with our custom Docker image",[12,1040,1041],{},"Now that we have built our Custom docker image, we can train the model without installing any more dependencies in the job.",[12,1043,1044,1045,1047],{},"The train job in our ",[62,1046,110],{}," looks like this:",[126,1049,1051],{"className":128,"code":1050,"language":130,"meta":131,"style":131},"train:\n  stage: train\n  tags:\n    - saas-linux-medium-amd64-gpu-standard\n  image: $IMAGE_PATH\n  script:\n    - python train_digit_recognizer.py\n  artifacts:\n    paths:\n      - mnist.h5\n    expose_as: 'trained model'\n\n",[62,1052,1053,1060,1068,1074,1080,1088,1094,1101,1108,1115,1123],{"__ignoreMap":131},[135,1054,1055,1058],{"class":137,"line":138},[135,1056,1057],{"class":141},"train",[135,1059,146],{"class":145},[135,1061,1062,1064,1066],{"class":137,"line":149},[135,1063,152],{"class":141},[135,1065,155],{"class":145},[135,1067,512],{"class":158},[135,1069,1070,1072],{"class":137,"line":162},[135,1071,165],{"class":141},[135,1073,146],{"class":145},[135,1075,1076,1078],{"class":137,"line":170},[135,1077,173],{"class":145},[135,1079,176],{"class":158},[135,1081,1082,1084,1086],{"class":137,"line":314},[135,1083,213],{"class":141},[135,1085,155],{"class":145},[135,1087,938],{"class":158},[135,1089,1090,1092],{"class":137,"line":323},[135,1091,247],{"class":141},[135,1093,146],{"class":145},[135,1095,1096,1098],{"class":137,"line":330},[135,1097,173],{"class":145},[135,1099,1100],{"class":158},"python train_digit_recognizer.py\n",[135,1102,1103,1106],{"class":137,"line":338},[135,1104,1105],{"class":141},"  artifacts",[135,1107,146],{"class":145},[135,1109,1110,1113],{"class":137,"line":349},[135,1111,1112],{"class":141},"    paths",[135,1114,146],{"class":145},[135,1116,1117,1120],{"class":137,"line":357},[135,1118,1119],{"class":145},"      - ",[135,1121,1122],{"class":158},"mnist.h5\n",[135,1124,1125,1128,1130],{"class":137,"line":365},[135,1126,1127],{"class":141},"    expose_as",[135,1129,155],{"class":145},[135,1131,1132],{"class":158},"'trained model'\n",[12,1134,859,1135,1137],{},[62,1136,862],{}," to see the job logs.",[12,1139,1140],{},[523,1141],{"alt":1142,"src":1143},"Train job logs","https://res.cloudinary.com/about-gitlab-com/image/upload/v1782397796/blog/Content%20Images/2023-07-06-gpu-enabled-runners-for-modelops/train-job-log.png",[12,1145,1146,1147,1149],{},"From here, you can also inspect the ",[62,1148,1057],{}," job artifacts.",[528,1151,1153],{"id":1152},"publishing-the-model","Publishing the model",[12,1155,1156,1157,1159],{},"In the last step of our ",[62,1158,110],{}," file, we are going to publish the trained model.",[126,1161,1163],{"className":128,"code":1162,"language":130,"meta":131,"style":131},"publish:\n  stage: publish\n  when: manual\n  dependencies:\n    - train\n  image: curlimages/curl:latest\n  script:\n    - 'curl --header \"JOB-TOKEN: $CI_JOB_TOKEN\" --upload-file mnist.h5 \"${CI_API_V4_URL}/projects/${CI_PROJECT_ID}/packages/generic/MNIST-Model/${CI_COMMIT_TAG}/mnist.h5\"'\n\n",[62,1164,1165,1172,1180,1190,1197,1203,1212,1218],{"__ignoreMap":131},[135,1166,1167,1170],{"class":137,"line":138},[135,1168,1169],{"class":141},"publish",[135,1171,146],{"class":145},[135,1173,1174,1176,1178],{"class":137,"line":149},[135,1175,152],{"class":141},[135,1177,155],{"class":145},[135,1179,519],{"class":158},[135,1181,1182,1185,1187],{"class":137,"line":162},[135,1183,1184],{"class":141},"  when",[135,1186,155],{"class":145},[135,1188,1189],{"class":158},"manual\n",[135,1191,1192,1195],{"class":137,"line":170},[135,1193,1194],{"class":141},"  dependencies",[135,1196,146],{"class":145},[135,1198,1199,1201],{"class":137,"line":314},[135,1200,173],{"class":145},[135,1202,512],{"class":158},[135,1204,1205,1207,1209],{"class":137,"line":323},[135,1206,213],{"class":141},[135,1208,155],{"class":145},[135,1210,1211],{"class":158},"curlimages/curl:latest\n",[135,1213,1214,1216],{"class":137,"line":330},[135,1215,247],{"class":141},[135,1217,146],{"class":145},[135,1219,1220,1222],{"class":137,"line":338},[135,1221,173],{"class":145},[135,1223,1224],{"class":158},"'curl --header \"JOB-TOKEN: $CI_JOB_TOKEN\" --upload-file mnist.h5 \"${CI_API_V4_URL}/projects/${CI_PROJECT_ID}/packages/generic/MNIST-Model/${CI_COMMIT_TAG}/mnist.h5\"'\n",[12,1226,859,1227,1229,1230,1232,1233,1236],{},[62,1228,862],{}," and trigger the ",[62,1231,1169],{}," job manually. After that, navigate into ",[62,1234,1235],{},"Deploy > Package Registry"," to verify the uploaded trained model.",[12,1238,1239],{},[523,1240],{"alt":1241,"src":1242},"Package Registry","https://res.cloudinary.com/about-gitlab-com/image/upload/v1782397795/blog/Content%20Images/2023-07-06-gpu-enabled-runners-for-modelops/package-registry.png",[80,1244,1246],{"id":1245},"example-4-jupyter-notebook-model-training-for-ml-powered-gitlab-issue-triage","Example 4: Jupyter notebook model training for ML-powered GitLab Issue triage",[12,1248,1249,1250,1255],{},"In the last example, we are using our GPU-enabled runner to train the internal ",[19,1251,1254],{"href":1252,"rel":1253},"https://gitlab.com/gitlab-org/ml-ops/tanuki-stan/-/tree/using-gpu-enabled-runner",[],"GitLab model to triage issues",". We use this model at GitLab to determine and assign issues to the right team from the context of the issue description.",[12,1257,1258,1259,1267,1268,1273],{},"Different from the previous examples, we now use the ",[19,1260,1263,1266],{"href":1261,"rel":1262},"https://hub.docker.com/r/tensorflow/tensorflow",[],[62,1264,1265],{},"tensorflow-gpu"," container image"," and install the ",[19,1269,1272],{"href":1270,"rel":1271},"https://gitlab.com/gitlab-org/ml-ops/tanuki-stan/-/blob/using-gpu-enabled-runner/notebooks/requirements.tensorflow-gpu.txt",[],"requirements"," in the job itself.",[12,1275,1276,1278],{},[62,1277,110],{}," configuration:",[126,1280,1282],{"className":128,"code":1281,"language":130,"meta":131,"style":131},"train:\n  tags:\n    - saas-linux-medium-amd64-gpu-standard\n  image: tensorflow/tensorflow:2.4.1-gpu\n  script:\n    - nvidia-smi\n    - cd notebooks\n    - pip install -r requirements.tensorflow-gpu.txt\n    - jupyter nbconvert --to script classify_groups.ipynb\n    - apt-get install -y p7zip-full\n    - cd ../data\n    - 7z x -p${DATA_PASSWORD} gitlab-issues.7z\n    - cd ../notebooks\n    - python3 classify_groups.py\n  artifacts:\n    paths:\n      - models/\n  rules:\n    - if: $CI_PIPELINE_SOURCE == \"merge_request_event\" || $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH  \n      when: manual\n      allow_failure: true\n\n",[62,1283,1284,1290,1296,1302,1311,1317,1323,1330,1337,1344,1351,1358,1365,1372,1379,1385,1391,1398,1404,1418,1428],{"__ignoreMap":131},[135,1285,1286,1288],{"class":137,"line":138},[135,1287,1057],{"class":141},[135,1289,146],{"class":145},[135,1291,1292,1294],{"class":137,"line":149},[135,1293,165],{"class":141},[135,1295,146],{"class":145},[135,1297,1298,1300],{"class":137,"line":162},[135,1299,173],{"class":145},[135,1301,176],{"class":158},[135,1303,1304,1306,1308],{"class":137,"line":170},[135,1305,213],{"class":141},[135,1307,155],{"class":145},[135,1309,1310],{"class":158},"tensorflow/tensorflow:2.4.1-gpu\n",[135,1312,1313,1315],{"class":137,"line":314},[135,1314,247],{"class":141},[135,1316,146],{"class":145},[135,1318,1319,1321],{"class":137,"line":323},[135,1320,173],{"class":145},[135,1322,256],{"class":158},[135,1324,1325,1327],{"class":137,"line":330},[135,1326,173],{"class":145},[135,1328,1329],{"class":158},"cd notebooks\n",[135,1331,1332,1334],{"class":137,"line":338},[135,1333,173],{"class":145},[135,1335,1336],{"class":158},"pip install -r requirements.tensorflow-gpu.txt\n",[135,1338,1339,1341],{"class":137,"line":349},[135,1340,173],{"class":145},[135,1342,1343],{"class":158},"jupyter nbconvert --to script classify_groups.ipynb\n",[135,1345,1346,1348],{"class":137,"line":357},[135,1347,173],{"class":145},[135,1349,1350],{"class":158},"apt-get install -y p7zip-full\n",[135,1352,1353,1355],{"class":137,"line":365},[135,1354,173],{"class":145},[135,1356,1357],{"class":158},"cd ../data\n",[135,1359,1360,1362],{"class":137,"line":621},[135,1361,173],{"class":145},[135,1363,1364],{"class":158},"7z x -p${DATA_PASSWORD} gitlab-issues.7z\n",[135,1366,1367,1369],{"class":137,"line":626},[135,1368,173],{"class":145},[135,1370,1371],{"class":158},"cd ../notebooks\n",[135,1373,1374,1376],{"class":137,"line":632},[135,1375,173],{"class":145},[135,1377,1378],{"class":158},"python3 classify_groups.py\n",[135,1380,1381,1383],{"class":137,"line":641},[135,1382,1105],{"class":141},[135,1384,146],{"class":145},[135,1386,1387,1389],{"class":137,"line":646},[135,1388,1112],{"class":141},[135,1390,146],{"class":145},[135,1392,1393,1395],{"class":137,"line":652},[135,1394,1119],{"class":145},[135,1396,1397],{"class":158},"models/\n",[135,1399,1400,1402],{"class":137,"line":804},[135,1401,807],{"class":141},[135,1403,146],{"class":145},[135,1405,1406,1408,1410,1412,1415],{"class":137,"line":812},[135,1407,173],{"class":145},[135,1409,817],{"class":141},[135,1411,155],{"class":145},[135,1413,1414],{"class":158},"$CI_PIPELINE_SOURCE == \"merge_request_event\" || $CI_COMMIT_BRANCH == $CI_DEFAULT_BRANCH",[135,1416,1417],{"class":145},"  \n",[135,1419,1421,1424,1426],{"class":137,"line":1420},20,[135,1422,1423],{"class":141},"      when",[135,1425,155],{"class":145},[135,1427,1189],{"class":158},[135,1429,1431,1434,1436],{"class":137,"line":1430},21,[135,1432,1433],{"class":141},"      allow_failure",[135,1435,155],{"class":145},[135,1437,1439],{"class":1438},"sYu0t","true\n",[12,1441,1442],{},[523,1443],{"alt":1444,"src":1445},"TensorFlow train","https://res.cloudinary.com/about-gitlab-com/image/upload/v1782397795/blog/Content%20Images/2023-07-06-gpu-enabled-runners-for-modelops/tensorflow-train.png",[12,1447,1448,1449,71],{},"If you are interested in another Jupyter notebook example, check out our recently published video on ",[19,1450,1453],{"href":1451,"rel":1452},"https://youtu.be/tElegG4NCZ0",[],"Training ML Models using GPU-enabled runner",[1455,1456],"iframe",{"width":1457,"height":1458,"src":1459,"title":1460,"frameBorder":1461,"allow":1462,"allowFullScreen":207},768,432,"https://www.youtube.com/embed/tElegG4NCZ0","YouTube video player","0","accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share",[48,1464,1466],{"id":1465},"results","Results",[12,1468,1469],{},"The integration of GPU-enabled runners on GitLab.com SaaS opens up a new realm of possibilities for ModelOps and HPC workloads.\nBy harnessing the power of GPU-enabled runners, you can accelerate your machine learning workflows, enable faster data processing, and improve scientific simulations, all while taking full advantage of a SaaS solution and avoiding the hurdles of hosting and maintaining your own build hardware.",[12,1471,1472,1473,71],{},"When you try the GPU-enabled runners, please share your experience in our ",[19,1474,1477],{"href":1475,"rel":1476},"https://gitlab.com/gitlab-org/gitlab/-/issues/403008",[],"feedback issue",[12,1479,1480,1481,1486,1487,1492],{},"Compute-heavy workloads can take a long time. A known problem is timeouts after three hours because of the current ",[19,1482,1485],{"href":1483,"rel":1484},"https://docs.gitlab.com/ci/runners/#how-saas-runners-work",[],"configuration of GitLab SaaS runners",".\nWe plan to release more powerful compute for future iterations to handle heavier workloads faster. You can follow updates about GPU-enabled runners in the ",[19,1488,1491],{"href":1489,"rel":1490},"https://gitlab.com/groups/gitlab-org/-/epics/8648",[],"GPU-enabled runners epic"," and learn more in the GPU-enabled runners documentation.",[1494,1495,1496],"style",{},"html pre.shiki code .shJU0, html code.shiki .shJU0{--shiki-default:#22863A}html pre.shiki code .sgsFI, html code.shiki .sgsFI{--shiki-default:#24292E}html pre.shiki code .sYBdl, html code.shiki .sYBdl{--shiki-default:#032F62}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html pre.shiki code .sYu0t, html code.shiki .sYu0t{--shiki-default:#005CC5}html pre.shiki code .sD7c4, html code.shiki .sD7c4{--shiki-default:#D73A49}",{"title":131,"searchDepth":149,"depth":149,"links":1498},[1499,1500,1506,1510,1514],{"id":50,"depth":149,"text":51},{"id":74,"depth":149,"text":75,"children":1501},[1502,1503,1504,1505],{"id":82,"depth":162,"text":83},{"id":103,"depth":162,"text":104},{"id":179,"depth":162,"text":180},{"id":221,"depth":162,"text":222},{"id":259,"depth":149,"text":260,"children":1507},[1508,1509],{"id":266,"depth":162,"text":267},{"id":373,"depth":162,"text":374},{"id":437,"depth":149,"text":438,"children":1511},[1512,1513],{"id":444,"depth":162,"text":445},{"id":1245,"depth":162,"text":1246},{"id":1465,"depth":149,"text":1466},"ai","2023-07-06","Learn how to leverage our GitLab-hosted GPU-enabled runners for ModelOps and high-performance computing workloads.","md",null,false,"https://res.cloudinary.com/about-gitlab-com/image/upload/v1749682702/Blog/Hero%20Images/gitlab-data-science-icon.png",{},"/en-us/blog/empowering-modelops-and-hpc-workloads-with-gpu-enabled-runners",{"title":1525,"description":1517,"ogTitle":1525,"ogDescription":1517,"noIndex":1520,"ogImage":1521,"ogUrl":1526,"ogSiteName":1527,"ogType":1528,"canonicalUrls":1526},"GPU-enabled runners for ModelOps and HPC workloads in CI/CD","https://about.gitlab.com/blog/empowering-modelops-and-hpc-workloads-with-gpu-enabled-runners","https://about.gitlab.com","article","empowering-modelops-and-hpc-workloads-with-gpu-enabled-runners","en-us/blog/empowering-modelops-and-hpc-workloads-with-gpu-enabled-runners",[1532,1533,1534,1535,1536],"DevSecOps","AI","CI/CD","product","features","BlogPost","Y5O4UXOcNRRsPoGlALI_iGQBYfEcDeTRTnY50azFIl4",{"logo":1540,"freeTrial":1545,"sales":1550,"login":1555,"items":1560,"search":1895,"minimal":1927,"duo":1946,"switchNav":1955,"pricingDeployment":1966},{"config":1541},{"href":1542,"dataGaName":1543,"dataGaLocation":1544},"/","gitlab logo","header",{"text":1546,"config":1547},"Get free trial",{"href":1548,"dataGaName":1549,"dataGaLocation":1544},"https://gitlab.com/-/trial_registrations/new?glm_source=about.gitlab.com&glm_content=default-saas-trial/","free trial",{"text":1551,"config":1552},"Request a demo",{"href":1553,"dataGaName":1554,"dataGaLocation":1544},"/request-a-demo/","sales",{"text":1556,"config":1557},"Sign in",{"href":1558,"dataGaName":1559,"dataGaLocation":1544},"https://gitlab.com/users/sign_in/","sign in",[1561,1646,1734,1739,1819,1873],{"text":1562,"left":207,"config":1563,"menu":1565},"Platform",{"dataNavLevelOne":1564},"platform",{"type":1566,"link":1567,"columns":1571,"feature":1635},"lists",{"text":1568,"config":1569},"Explore the Platform",{"href":1570,"dataGaName":1564,"dataGaLocation":1544},"/platform/",[1572,1594,1617],{"title":1573,"items":1574},"DevOps",[1575,1579,1584,1589],{"text":1534,"config":1576},{"href":1577,"dataGaLocation":1544,"dataGaName":1578},"/solutions/continuous-integration/","ci/cd",{"text":1580,"config":1581},"Source Code Management",{"href":1582,"dataGaLocation":1544,"dataGaName":1583},"/solutions/source-code-management/","source code management",{"text":1585,"config":1586},"Agile Delivery",{"href":1587,"dataGaLocation":1544,"dataGaName":1588},"/solutions/agile-delivery/","agile delivery",{"text":1590,"config":1591},"Artifact Management",{"href":1592,"dataGaLocation":1544,"dataGaName":1593},"/artifact-central/","artifact management",{"title":1595,"items":1596},"Security & Governance",[1597,1602,1607,1612],{"text":1598,"config":1599},"Application Security Testing",{"href":1600,"dataGaLocation":1544,"dataGaName":1601},"/solutions/application-security-testing/","application security testing",{"text":1603,"config":1604},"Governance & Compliance",{"href":1605,"dataGaLocation":1544,"dataGaName":1606},"/solutions/software-compliance/","governance and compliance",{"text":1608,"config":1609},"Secrets Management",{"href":1610,"dataGaLocation":1544,"dataGaName":1611},"/gitlab-secrets-manager/","secrets management",{"text":1613,"config":1614},"Supply Chain Security",{"href":1615,"dataGaLocation":1544,"dataGaName":1616},"/solutions/supply-chain/","supply chain security",{"title":1618,"items":1619},"Agentic AI & Context",[1620,1625,1630],{"text":1621,"config":1622},"Duo Agent Platform",{"href":1623,"dataGaLocation":1544,"dataGaName":1624},"/gitlab-duo-agent-platform/","duo agent platform",{"text":1626,"config":1627},"Orbit Context Graph",{"href":1628,"dataGaLocation":1544,"dataGaName":1629},"/gitlab-orbit/","orbit",{"text":1631,"config":1632},"Visibility & Measurement",{"href":1633,"dataGaLocation":1544,"dataGaName":1634},"/solutions/visibility-measurement/","visibility and measurement",{"config":1636,"title":1639,"text":1640,"link":1641},{"background":1637,"textColor":1638},"url('https://res.cloudinary.com/about-gitlab-com/image/upload/v1790185102/o7dn32anqcb6z8dgd3as.jpg')","#000","GitLab Orbit is generally available","Higher agent accuracy, lower token cost",{"text":1642,"config":1643},"Read the announcement",{"href":1644,"dataGaName":1645,"dataGaLocation":1544},"/blog/gitlab-orbit-ga/","orbit ga announcement",{"text":1647,"left":207,"config":1648,"menu":1650},"Solutions",{"dataNavLevelOne":1649},"solutions",{"type":1566,"link":1651,"columns":1655,"feature":1725},{"text":1652,"config":1653},"View all Solutions",{"href":1654,"dataGaName":1649,"dataGaLocation":1544},"/solutions/",[1656,1674,1692],{"title":1657,"items":1658},"Outcomes",[1659,1664,1669],{"text":1660,"config":1661},"DevOps Modernization",{"href":1662,"dataGaLocation":1544,"dataGaName":1663},"/assessments/devops-modernization-assessment/","devops modernization",{"text":1665,"config":1666},"Security Modernization",{"href":1667,"dataGaLocation":1544,"dataGaName":1668},"/assessments/security-modernization-assessment/","security modernization",{"text":1670,"config":1671},"AI Modernization",{"href":1672,"dataGaLocation":1544,"dataGaName":1673},"/assessments/ai-modernization-assessment/","ai modernization",{"title":1675,"items":1676},"By size",[1677,1682,1687],{"text":1678,"config":1679},"Enterprise",{"href":1680,"dataGaLocation":1544,"dataGaName":1681},"/enterprise/","enterprise",{"text":1683,"config":1684},"Small Business",{"href":1685,"dataGaLocation":1544,"dataGaName":1686},"/small-business/","small business",{"text":1688,"config":1689},"Startups",{"href":1690,"dataGaLocation":1544,"dataGaName":1691},"/solutions/startups/","startups",{"title":1693,"items":1694},"Industries",[1695,1700,1705,1710,1715,1720],{"text":1696,"config":1697},"Financial Services",{"href":1698,"dataGaLocation":1544,"dataGaName":1699},"/solutions/finance/","financial services",{"text":1701,"config":1702},"Public Sector",{"href":1703,"dataGaLocation":1544,"dataGaName":1704},"/solutions/public-sector/","public sector",{"text":1706,"config":1707},"Telecommunications",{"href":1708,"dataGaLocation":1544,"dataGaName":1709},"/solutions/telecommunications/","telecommunications",{"text":1711,"config":1712},"Automotive",{"href":1713,"dataGaLocation":1544,"dataGaName":1714},"/solutions/automotive/","automotive",{"text":1716,"config":1717},"Education",{"href":1718,"dataGaLocation":1544,"dataGaName":1719},"/solutions/education/","education",{"text":1721,"config":1722},"Aerospace",{"href":1723,"dataGaLocation":1544,"dataGaName":1724},"/solutions/aerospace/","aerospace",{"config":1726,"title":1727,"text":1728,"link":1729},{"background":1637,"textColor":1638},"GitLab Security Standard","Learn how to harden your software factory and let AI agents earn autonomy, without trading away security.",{"text":1730,"config":1731},"Explore the standard",{"href":1732,"dataGaName":1733,"dataGaLocation":1544},"/gitlab-security-standard/","gitlab security standard",{"text":1735,"config":1736},"Pricing",{"href":1737,"dataGaName":1738,"dataGaLocation":1544,"dataNavLevelOne":1738},"/pricing/","pricing",{"text":1740,"config":1741,"menu":1743},"Resources",{"dataNavLevelOne":1742},"resources",{"type":1566,"link":1744,"columns":1748,"feature":1810},{"text":1745,"config":1746},"View all resources",{"href":1747,"dataGaName":1742,"dataGaLocation":1544},"/resources/",[1749,1782],{"title":1750,"items":1751},"Discover",[1752,1757,1762,1767,1772,1777],{"text":1753,"config":1754},"Docs",{"href":1755,"dataGaName":1756,"dataGaLocation":1544},"https://docs.gitlab.com/","docs",{"text":1758,"config":1759},"University",{"href":1760,"dataGaName":1761,"dataGaLocation":1544},"https://university.gitlab.com/","university",{"text":1763,"config":1764},"Demo Series",{"href":1765,"dataGaName":1766,"dataGaLocation":1544},"/gitlab-technical-demo-series/","demo series",{"text":1768,"config":1769},"Demo Hub",{"href":1770,"dataGaName":1771,"dataGaLocation":1544},"/demo-hub/","demo hub",{"text":1773,"config":1774},"Services",{"href":1775,"dataGaName":1776,"dataGaLocation":1544},"/services/","services",{"text":1778,"config":1779},"Well Architected",{"href":1780,"dataGaName":1781,"dataGaLocation":1544},"/resources/well-architected/","well architected",{"title":1783,"items":1784},"Connect",[1785,1790,1795,1800,1805],{"text":1786,"config":1787},"Blog",{"href":1788,"dataGaName":1789,"dataGaLocation":1544},"/blog/","blog",{"text":1791,"config":1792},"Community",{"href":1793,"dataGaName":1794,"dataGaLocation":1544},"/community/","community",{"text":1796,"config":1797},"Customers",{"href":1798,"dataGaName":1799,"dataGaLocation":1544},"/customers/","customers",{"text":1801,"config":1802},"Partners",{"href":1803,"dataGaName":1804,"dataGaLocation":1544},"/partners/","partners",{"text":1806,"config":1807},"Events",{"href":1808,"dataGaName":1809,"dataGaLocation":1544},"/events/","events",{"config":1811,"title":1812,"text":1813,"link":1814},{"background":1637,"textColor":1638},"What’s new in GitLab","Stay updated with our latest features and improvements.",{"text":1815,"config":1816},"Read the latest",{"href":1817,"dataGaName":1818,"dataGaLocation":1544},"/whats-new/","whats new",{"text":1820,"config":1821,"menu":1823},"Company",{"dataNavLevelOne":1822},"company",{"type":1566,"columns":1824},[1825],{"items":1826},[1827,1832,1838,1843,1848,1853,1858,1863,1868],{"text":1828,"config":1829},"About",{"href":1830,"dataGaName":1831,"dataGaLocation":1544},"/company/","about",{"text":1833,"config":1834,"footerGa":1837},"Jobs",{"href":1835,"dataGaName":1836,"dataGaLocation":1544},"/jobs/","jobs",{"dataGaName":1836},{"text":1839,"config":1840},"Press",{"href":1841,"dataGaName":1842,"dataGaLocation":1544},"/press/","press",{"text":1844,"config":1845},"Handbook",{"href":1846,"dataGaName":1847,"dataGaLocation":1544},"https://handbook.gitlab.com/","handbook",{"text":1849,"config":1850},"Leadership",{"href":1851,"dataGaName":1852,"dataGaLocation":1544},"/company/team/e-group/","leadership",{"text":1854,"config":1855},"Investor relations",{"href":1856,"dataGaName":1857,"dataGaLocation":1544},"https://ir.gitlab.com/overview/default.aspx","investor relations",{"text":1859,"config":1860},"Trust Center",{"href":1861,"dataGaName":1862,"dataGaLocation":1544},"/security/","trust center",{"text":1864,"config":1865},"AI Transparency Center",{"href":1866,"dataGaName":1867,"dataGaLocation":1544},"/ai-transparency-center/","ai transparency center",{"text":1869,"config":1870},"Newsletter",{"href":1871,"dataGaName":1872,"dataGaLocation":1544},"/company/contact/#contact-forms","newsletter",{"text":1874,"config":1875,"menu":1876},"Contact us",{"dataNavLevelOne":1822},{"type":1566,"columns":1877},[1878],{"items":1879},[1880,1885,1890],{"text":1881,"config":1882},"Talk to sales",{"href":1883,"dataGaName":1884,"dataGaLocation":1544},"/sales/","talk to sales",{"text":1886,"config":1887},"Support portal",{"href":1888,"dataGaName":1889,"dataGaLocation":1544},"https://support.gitlab.com/hc/en-us","support portal",{"text":1891,"config":1892},"Customer portal",{"href":1893,"dataGaName":1894,"dataGaLocation":1544},"https://customers.gitlab.com/customers/sign_in/","customer portal",{"close":1896,"login":1897,"suggestions":1904},"Close",{"text":1898,"link":1899},"To search repositories and projects, login to",{"text":1900,"config":1901},"gitlab.com",{"href":1558,"dataGaName":1902,"dataGaLocation":1903},"search login","search",{"text":1905,"default":1906},"Suggestions",[1907,1910,1913,1915,1919,1923],{"text":1908,"config":1909},"GitLab Duo Agent Platform",{"href":1623,"dataGaName":1908,"dataGaLocation":1903},{"text":1911,"config":1912},"Code Suggestions (AI)",{"href":1623,"dataGaName":1911,"dataGaLocation":1903},{"text":1534,"config":1914},{"href":1577,"dataGaName":1534,"dataGaLocation":1903},{"text":1916,"config":1917},"GitLab on AWS",{"href":1918,"dataGaName":1916,"dataGaLocation":1903},"/partners/technology-partners/aws/",{"text":1920,"config":1921},"GitLab on Google Cloud",{"href":1922,"dataGaName":1920,"dataGaLocation":1903},"/partners/technology-partners/google-cloud-platform/",{"text":1924,"config":1925},"Why GitLab?",{"href":1926,"dataGaName":1924,"dataGaLocation":1903},"/why-gitlab/",{"freeTrial":1928,"mobileIcon":1933,"desktopIcon":1938,"secondaryButton":1941},{"text":1929,"config":1930},"Start free trial",{"href":1931,"dataGaName":1549,"dataGaLocation":1932},"https://gitlab.com/-/trials/new/","nav",{"altText":1934,"config":1935},"Gitlab Icon",{"src":1936,"dataGaName":1937,"dataGaLocation":1932},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1758203874/jypbw1jx72aexsoohd7x.svg","gitlab icon",{"altText":1934,"config":1939},{"src":1940,"dataGaName":1937,"dataGaLocation":1932},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1758203875/gs4c8p8opsgvflgkswz9.svg",{"text":1942,"config":1943},"Get Started",{"href":1944,"dataGaName":1945,"dataGaLocation":1932},"https://gitlab.com/-/trial_registrations/new?glm_source=about.gitlab.com/get-started/","get started",{"freeTrial":1947,"mobileIcon":1951,"desktopIcon":1953},{"text":1948,"config":1949},"Learn more about GitLab Duo",{"href":1623,"dataGaName":1950,"dataGaLocation":1932},"gitlab duo",{"altText":1934,"config":1952},{"src":1936,"dataGaName":1937,"dataGaLocation":1932},{"altText":1934,"config":1954},{"src":1940,"dataGaName":1937,"dataGaLocation":1932},{"button":1956,"mobileIcon":1961,"desktopIcon":1963},{"text":1957,"config":1958},"/switch",{"href":1959,"dataGaName":1960,"dataGaLocation":1932},"#contact","switch",{"altText":1934,"config":1962},{"src":1936,"dataGaName":1937,"dataGaLocation":1932},{"altText":1934,"config":1964},{"src":1965,"dataGaName":1937,"dataGaLocation":1932},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1773335277/ohhpiuoxoldryzrnhfrh.png",{"freeTrial":1967,"mobileIcon":1972,"desktopIcon":1974},{"text":1968,"config":1969},"Back to pricing",{"href":1737,"dataGaName":1970,"dataGaLocation":1932,"icon":1971},"back to pricing","GoBack",{"altText":1934,"config":1973},{"src":1936,"dataGaName":1937,"dataGaLocation":1932},{"altText":1934,"config":1975},{"src":1940,"dataGaName":1937,"dataGaLocation":1932},{"title":1977,"button":1978,"config":1982},"GitLab Orbit is generally available. Increase agent accuracy and lower token cost",{"text":1642,"config":1979},{"href":1644,"dataGaName":1980,"dataGaLocation":1981},"orbit-ga-banner","global-banner",{"layout":1983,"disabled":1520},"release",{"data":1985},{"text":1986,"config":1987,"items":1992,"minimal":2219},"Git is a trademark of Software Freedom Conservancy and our use of 'GitLab' is under license",{"twitter":1988,"facebook":1989,"youtube":1990,"linkedin":1991},"https://twitter.com/gitlab","https://www.facebook.com/gitlab","https://www.youtube.com/channel/UCnMGQ8QHMAnVIsI3xJrihhg","https://www.linkedin.com/company/gitlab-com",[1993,2041,2099,2146,2187],{"title":1735,"links":1994,"subMenu":2010},[1995,2000,2005],{"text":1996,"config":1997},"View plans",{"href":1737,"dataGaName":1998,"dataGaLocation":1999},"view plans","footer",{"text":2001,"config":2002},"Why Premium?",{"href":2003,"dataGaName":2004,"dataGaLocation":1999},"/pricing/premium/","why premium",{"text":2006,"config":2007},"Why Ultimate?",{"href":2008,"dataGaName":2009,"dataGaLocation":1999},"/pricing/ultimate/","why ultimate",[2011],{"title":2012,"links":2013},"Contact Us",[2014,2017,2019,2021,2026,2031,2036],{"text":2015,"config":2016},"Contact sales",{"href":1883,"dataGaName":1554,"dataGaLocation":1999},{"text":1886,"config":2018},{"href":1888,"dataGaName":1889,"dataGaLocation":1999},{"text":1891,"config":2020},{"href":1893,"dataGaName":1894,"dataGaLocation":1999},{"text":2022,"config":2023},"Status",{"href":2024,"dataGaName":2025,"dataGaLocation":1999},"https://status.gitlab.com/","status",{"text":2027,"config":2028},"Terms of use",{"href":2029,"dataGaName":2030,"dataGaLocation":1999},"/terms/","terms of use",{"text":2032,"config":2033},"Privacy statement",{"href":2034,"dataGaName":2035,"dataGaLocation":1999},"/privacy/","privacy statement",{"text":2037,"config":2038},"Cookie preferences",{"dataGaName":2039,"dataGaLocation":1999,"id":2040,"isOneTrustButton":207},"cookie preferences","ot-sdk-btn",{"title":1562,"links":2042,"subMenu":2058},[2043,2046,2048,2050,2052,2054],{"text":2044,"config":2045},"Explore the platform",{"href":1570,"dataGaName":1564,"dataGaLocation":1999},{"text":1621,"config":2047},{"href":1623,"dataGaName":1624,"dataGaLocation":1999},{"text":1534,"config":2049},{"href":1577,"dataGaName":1578,"dataGaLocation":1999},{"text":1626,"config":2051},{"href":1628,"dataGaName":1629,"dataGaLocation":1999},{"text":1608,"config":2053},{"href":1610,"dataGaName":1611,"dataGaLocation":1999},{"text":2055,"config":2056},"Why GitLab",{"href":1926,"dataGaName":2057,"dataGaLocation":1999},"why gitlab",[2059],{"title":2060,"links":2061},"Topics",[2062,2066,2071,2075,2080,2084,2089,2094],{"text":1534,"config":2063},{"href":2064,"dataGaName":2065,"dataGaLocation":1999},"/topics/ci-cd/","cicd",{"text":2067,"config":2068},"GitOps",{"href":2069,"dataGaName":2070,"dataGaLocation":1999},"/topics/gitops/","gitops",{"text":1573,"config":2072},{"href":2073,"dataGaName":2074,"dataGaLocation":1999},"/topics/devops/","devops",{"text":2076,"config":2077},"Version Control",{"href":2078,"dataGaName":2079,"dataGaLocation":1999},"/topics/version-control/","version control",{"text":1532,"config":2081},{"href":2082,"dataGaName":2083,"dataGaLocation":1999},"/topics/devsecops/","devsecops",{"text":2085,"config":2086},"Cloud Native",{"href":2087,"dataGaName":2088,"dataGaLocation":1999},"/topics/cloud-native/","cloud native",{"text":2090,"config":2091},"AI for Coding",{"href":2092,"dataGaName":2093,"dataGaLocation":1999},"/topics/devops/ai-for-coding/","ai for coding",{"text":2095,"config":2096},"Agentic AI",{"href":2097,"dataGaName":2098,"dataGaLocation":1999},"/topics/agentic-ai/","agentic ai",{"title":1647,"links":2100},[2101,2104,2107,2110,2113,2116,2118,2120,2122,2124,2126,2129,2131,2134,2138,2142],{"text":2102,"config":2103},"DevOps modernization",{"href":1662,"dataGaName":1663,"dataGaLocation":1999},{"text":2105,"config":2106},"Security modernization",{"href":1667,"dataGaName":1668,"dataGaLocation":1999},{"text":2108,"config":2109},"AI modernization",{"href":1672,"dataGaName":1673,"dataGaLocation":1999},{"text":2111,"config":2112},"Financial services",{"href":1698,"dataGaName":1699,"dataGaLocation":1999},{"text":2114,"config":2115},"Public sector",{"href":1703,"dataGaName":1704,"dataGaLocation":1999},{"text":1706,"config":2117},{"href":1708,"dataGaName":1709,"dataGaLocation":1999},{"text":1711,"config":2119},{"href":1713,"dataGaName":1714,"dataGaLocation":1999},{"text":1716,"config":2121},{"href":1718,"dataGaName":1719,"dataGaLocation":1999},{"text":1721,"config":2123},{"href":1723,"dataGaName":1724,"dataGaLocation":1999},{"text":1678,"config":2125},{"href":1680,"dataGaName":1681,"dataGaLocation":1999},{"text":2127,"config":2128},"Small business",{"href":1685,"dataGaName":1686,"dataGaLocation":1999},{"text":1688,"config":2130},{"href":1690,"dataGaName":1691,"dataGaLocation":1999},{"text":2067,"config":2132},{"href":2133,"dataGaName":2070,"dataGaLocation":1999},"/solutions/gitops/",{"text":2135,"config":2136},"Software composition analysis",{"href":1615,"dataGaName":2137,"dataGaLocation":1999},"software composition analysis",{"text":2139,"config":2140},"Value stream management",{"href":1633,"dataGaName":2141,"dataGaLocation":1999},"value stream management",{"text":2143,"config":2144},"All solutions",{"href":1654,"dataGaName":2145,"dataGaLocation":1999},"all solutions",{"title":1740,"links":2147},[2148,2153,2158,2160,2162,2164,2166,2168,2170,2172,2174,2176,2178,2180,2183],{"text":2149,"config":2150},"Install",{"href":2151,"dataGaName":2152,"dataGaLocation":1999},"/install/","install",{"text":2154,"config":2155},"Quick start guides",{"href":2156,"dataGaName":2157,"dataGaLocation":1999},"/get-started/","quick setup checklists",{"text":1753,"config":2159},{"href":1755,"dataGaName":1756,"dataGaLocation":1999},{"text":1758,"config":2161},{"href":1760,"dataGaName":1761,"dataGaLocation":1999},{"text":1763,"config":2163},{"href":1765,"dataGaName":1766,"dataGaLocation":1999},{"text":1768,"config":2165},{"href":1770,"dataGaName":1771,"dataGaLocation":1999},{"text":1773,"config":2167},{"href":1775,"dataGaName":1776,"dataGaLocation":1999},{"text":1786,"config":2169},{"href":1788,"dataGaName":1789,"dataGaLocation":1999},{"text":1791,"config":2171},{"href":1793,"dataGaName":1794,"dataGaLocation":1999},{"text":1796,"config":2173},{"href":1798,"dataGaName":1799,"dataGaLocation":1999},{"text":1801,"config":2175},{"href":1803,"dataGaName":1804,"dataGaLocation":1999},{"text":1806,"config":2177},{"href":1808,"dataGaName":1809,"dataGaLocation":1999},{"text":1869,"config":2179},{"href":1871,"dataGaName":1872,"dataGaLocation":1999},{"text":2181,"config":2182},"What's new",{"href":1817,"dataGaName":1818,"dataGaLocation":1999},{"text":2184,"config":2185},"All resources",{"href":1747,"dataGaName":2186,"dataGaLocation":1999},"all resources",{"title":1820,"links":2188},[2189,2191,2193,2195,2197,2199,2201,2203,2205,2209,2214],{"text":1828,"config":2190},{"href":1830,"dataGaName":1822,"dataGaLocation":1999},{"text":1833,"config":2192},{"href":1835,"dataGaName":1836,"dataGaLocation":1999},{"text":1839,"config":2194},{"href":1841,"dataGaName":1842,"dataGaLocation":1999},{"text":1844,"config":2196},{"href":1846,"dataGaName":1847,"dataGaLocation":1999},{"text":1849,"config":2198},{"href":1851,"dataGaName":1852,"dataGaLocation":1999},{"text":1854,"config":2200},{"href":1856,"dataGaName":1857,"dataGaLocation":1999},{"text":1859,"config":2202},{"href":1861,"dataGaName":1862,"dataGaLocation":1999},{"text":1864,"config":2204},{"href":1866,"dataGaName":1867,"dataGaLocation":1999},{"text":2206,"config":2207},"Sustainability",{"href":2208,"dataGaName":2206,"dataGaLocation":1999},"/sustainability/",{"text":2210,"config":2211},"Diversity, inclusion and belonging (DIB)",{"href":2212,"dataGaName":2213,"dataGaLocation":1999},"/diversity-inclusion-belonging/","Diversity, inclusion and belonging",{"text":2215,"config":2216},"Modern Slavery Transparency Statement",{"href":2217,"dataGaName":2218,"dataGaLocation":1999},"https://handbook.gitlab.com/handbook/legal/modern-slavery-act-transparency-statement/","modern slavery transparency statement",{"items":2220},[2221,2224,2227],{"text":2222,"config":2223},"Terms",{"href":2029,"dataGaName":2030,"dataGaLocation":1999},{"text":2225,"config":2226},"Cookies",{"dataGaName":2039,"dataGaLocation":1999,"id":2040,"isOneTrustButton":207},{"text":2228,"config":2229},"Privacy",{"href":2034,"dataGaName":2035,"dataGaLocation":1999},[2231],{"id":2232,"title":7,"body":1519,"config":2233,"content":2235,"description":1519,"extension":2239,"meta":2240,"navigation":207,"path":2241,"seo":2242,"stem":2243,"__hash__":2244},"blogAuthors/en-us/blog/authors/gabriel-engel.yml",{"template":2234},"BlogAuthor",{"name":7,"config":2236},{"headshot":2237,"ctfId":2238},"https://res.cloudinary.com/about-gitlab-com/image/upload/v1749664747/Blog/Author%20Headshots/gabrielengel_gl-headshot.jpg","gabrielengelgl","yml",{},"/en-us/blog/authors/gabriel-engel",{},"en-us/blog/authors/gabriel-engel","iVkglJtikdZz4_wER91EaHRwdIt1ZjN3O1LGKFcDThM",[2246,2254,2262],{"title":2247,"description":2248,"heroImage":2249,"category":1515,"date":2250,"authors":2251,"slug":2253,"externalUrl":1519},"Turn Jira tickets into GitLab merge requests with Duo Agent Platform","Learn how to build a custom Jira agent with GitLab Duo Agent Platform to manage your tickets and turn them into merge requests.","https://res.cloudinary.com/about-gitlab-com/image/upload/v1774028474/bvfgeoiz0pijij3ow7vd.png","2026-11-12",[2252],"Itzik Gan Baruch","jira-ticket-to-gitlab-mr-with-duo-agent-platform",{"title":2255,"description":2256,"heroImage":2257,"category":1515,"date":2258,"authors":2259,"slug":2261,"externalUrl":1519},"How to design a decision-grade pilot for agentic AI","A practical framework for deciding when agentic AI can safely act in your software lifecycle, using workflow evidence, risk-based thresholds, and governance.","https://res.cloudinary.com/about-gitlab-com/image/upload/v1774957323/xfnzvlwc4c2vhal3v3jv.png","2026-11-09",[2260],"Nina Sesti","how-to-develop-a-decision-grade-ai-pilot",{"title":2263,"description":2264,"heroImage":2265,"category":1515,"date":2266,"authors":2267,"slug":2269,"externalUrl":1519},"The Pentagon wants to know who checks code when AI writes it","Learn what Department of War Instruction 8430.01 asks of AI-assisted software delivery, and how GitLab keeps AI code traceable and verified.","https://res.cloudinary.com/about-gitlab-com/image/upload/v1776347152/unw3mzatkd5xyfbzcnni.png","2026-11-06",[2268],"Elisabeth Burrows","department-of-war-instruction-for-ai-generated-code",{"header":2271,"text":2272,"button":2276,"disclaimer":2279},"Start your free\n\n30-day GitLab trial\n",[2273,2274,2275],"GitLab Duo Agent Platform access","Try our most advanced features","Automate complex tasks with AI",{"text":1546,"config":2277},{"href":2278,"dataGaName":1549,"dataGaLocation":1789},"https://gitlab.com/-/trials/new?glm_content=default-saas-trial&glm_source=about.gitlab.com/blog/","No credit card required.",{"header":2281,"blurb":2282,"button":2283,"secondaryButton":2288},"Start building faster today","See what your team can do with the intelligent orchestration platform for DevSecOps.\n",{"text":2284,"config":2285},"Get your free trial",{"href":2286,"dataGaName":1549,"dataGaLocation":2287},"https://gitlab.com/-/trial_registrations/new?glm_content=default-saas-trial&glm_source=about.gitlab.com/","feature",{"text":2015,"config":2289},{"href":1883,"dataGaName":1554,"dataGaLocation":2287},1794911789895]