Help Center/ ModelArts/ Drawer/ Creating a Model/ Meta Model Source
Updated on 2025-05-29 GMT+08:00
Share

Meta Model Source

Description

When creating a model, you need to set different parameters for meta models from different sources.

  • Training job

    Create a training job on ModelArts and train a model. After obtaining your desired model, import and use it for deployment.

  • Container image

    For AI engines that are not supported by ModelArts, you can import the models to ModelArts from model images. ModelArts supports hitless rolling upgrade of real-time services through heal check configurations.

  • OBS

    If you use a frequently-used engine to develop and train a model locally, you can upload the model to OBS for model deployment. For details about the AI engines and their runtime environments supported by ModelArts preset images, see Supported AI Engines for ModelArts Inference.

Differences Between Model Sources

Source

Whether to Compile the Configuration File

Whether to Sett the Model/Image Storage Path

Whether to Support Non-typical AI Engines

Training job

x

x

x

Container image

x

√

√

OBS

√ (Optional)

√

x

Note: √ indicates required/supported, and x indicates not required/not supported.

FAQs

What Are the Specifications for Importing a Meta Model from OBS?

When creating a model on the model management page, make sure that any meta model imported from OBS complies with certain specifications.

  • The model package must contain the model folder, which stores the model file, model configuration file, and model inference code.
  • The model file must exist.
  • The model configuration file must exist and its name is fixed to config.json. There exists only one model configuration file. For details about how to write the model configuration file, see Specifications for Editing a Model Configuration File.
  • The model inference code file is required, the file name is fixed to customize_service.py. There must be only one such file. For details about how to write model inference code, see Specifications for Writing Model Inference Code.

Example of a TensorFlow-based model package imported from OBS

When publishing the model, you only need to specify the ocr directory.

OBS bucket or directory name
|── ocr
|   ├── model (Mandatory) Name of a fixed subdirectory, which is used to store model-related files
|   │   ├── <<Custom Python package>> (Optional) User's Python package, which can be directly referenced in model inference code
|   │   ├── saved_model.pb (Mandatory) Protocol buffer file, which contains the diagram description of the model
|   │   ├── variables Name of a fixed sub-directory, which contains the weight and deviation rate of the model. It is mandatory for the main file of the *.pb model.
|   │   │   ├── variables.index Mandatory
|   │   │   ├── variables.data-00000-of-00001 Mandatory
|   │   ├──config.json (Mandatory) Model configuration file. The file name is fixed to config.json. Only one model configuration file is supported.
|   │   ├──customize_service.py (Mandatory) Model inference code. The file name is fixed to customize_service.py. Only one model inference code file exists. The files on which customize_service.py depends can be directly stored in the model directory.

For more examples, see Model Package Example.

How Do I Create and Upload a Model Image for Importing a Meta Model From a Container Image?

When building a custom image using a locally developed model, ensure that the image complies with ModelArts specifications. For details, see Specifications for Custom Images.

After creating an image, you need to upload the image to SWR. SWR provides easy-to-use, secure, and reliable image management throughout the image lifecycle.

For details, see Creating a Custom Image for Inference.

Is There a Limit on the Size of an Imported Meta Model?

ModelArts uses containers for deploying services. There are size limitations during container runtime. If the size of your model file, custom file, or system file exceeds the Docker size, a message will be displayed, indicating that the image space is insufficient.

Currently, the maximum Docker size for a container in a public resource pool is 10 GB, and the maximum Docker size for a container in a dedicated resource pool is 30 GB.

If the meta model is imported from OBS or training jobs, the total size of the base image, model file, code, data file, and installed software packages is included.

If the meta model is imported from a custom image, the total size of the decompressed image and the downloaded image files is included.

Parent Topic: Creating a Model