Skip to Content
briefguide

Quick Start

Prerequisites

  1. You need to register as a XXXCloud user and complete the real-name authentication.
  2. UModelVerse is currently in the invitation testing phase. If you want to experience the features, please contact your account manager or technical support to apply for functional permissions.

Steps

Only four steps are needed to complete the deployment of industry models: Data Import -> Model Fine-tuning -> Model Release -> Service Deployment

Step 1: Data Import

  1. Click on the “US3 Object Storage” product — Create storage space — File management, and store the data or models needed for training to US3. For specific operations, see the US3 Operation Document.

  2. Create a US3 token: XXXCloud Documentation Center.

    ?> Note: The token permissions need to include download, file listing, upload, and delete; currently, only the North China 2 storage space is supported.

  3. Click on the “UModelVerse Model Service Platform” product — Function menu “Dataset Management” — Create Dataset

  4. Select the training data imported into US3 for use during fine-tuning.

Step 2: Model Fine-tuning

Through model fine-tuning, the model can be combined with the training dataset, optimizing the model’s professionalism in industry and business by setting parameters, making it a key step in training industry models.

!> Currently, only SFT training is supported.

  1. Click on the “UModelVerse Model Service Platform” product — Function menu “Model Fine-tuning” — Create Task

  2. Complete the corresponding parameter configuration following the step-by-step prompts:

    1. Model Selection: Choose a pre-configured open-source model or a private model uploaded to US3

    2. Data Configuration: Select the training dataset created in the first step

    3. Parameter Configuration: Configure the training parameters

    4. Output Configuration: Configure the export path after model training completion

    5. Confirmation Configuration: Confirm all parameters and billing information, then issue the task

  3. Once confirmed, the task will be automatically dispatched and compute power will be called for model training. When the status changes to “Completed,” it indicates the training task has ended, and you can view the model details. During training, you can also check the monitoring chart at any time to observe the training process. If there are any anomalies, you can stop the training task at any time.

Step 3: Model Release

A model that has performed as expected in training can be released. Once released successfully, it will enter the “Model Management” interface, facilitating management of each model version and selection of specific models for subsequent evaluation.

Step 4: Model Evaluation

  1. Click on the “UModelVerse Model Service Platform” product — Function menu “Model Evaluation” — Create Evaluation Task
  2. Evaluation Mode: The default current model evaluation method is “Baseline Evaluation.”
  3. Select Dataset: Choose “Preset Dataset” or “Custom Dataset” (datasets published in dataset management).
  4. Evaluate Model: Select “Baseline Model” (mandatory) and “Comparison Model.”
  5. Stopword List: Choose “Preset Stopword List” or upload “Custom Stopword List” (applicable for custom datasets). If you choose “Preset Dataset,” skip this step.
  6. Storage Settings: Choose storage space through the authorization token; currently, only the North China 2 region is supported.

Once you create an evaluation task, the evaluation will proceed automatically. When the status changes to “Completed,” it means the evaluation task has ended, and you can view the evaluation results. Currently supported evaluation results are divided into two types:

  1. Preset Dataset Evaluation Report: Supports viewing evaluation reports. Metrics scoring (STEM, Social Sciences, Humanities, Other, Average) and result visualization.

  2. Custom Dataset Evaluation Report: Supports viewing evaluation reports and evaluation details. Metrics scoring (ROUGE-1, ROUGE-2, ROUGE-L, BLEU), result visualization, and evaluation details.

Step 5: Service Deployment

  1. Click on the “UModelVerse Model Service Platform” product — Function menu “Service Management” — Create Service

  2. Choose a preset model or my models (models published in model management)

  3. Select the number of resource replicas, and confirm billing information to create the service

​ The newly created service will automatically enter the deployment status. Once the deployment is successful, an API Key will be generated, which can be used for service invocation.

​ You can also perform online access to the successfully deployed service and verify the service through conversations.

APIprivate