A model is a core object in MaxCompute. MaxCompute supports multiple model types, such as public, imported, and remote models, and provides unified management for models and model versions. This helps you easily integrate model capabilities into your business analytics workflows. This topic describes the basic concepts, benefits, and types of MaxCompute models and explains how to manage and use them.
Introduction
Concepts
Model: A model is a deployment object registered in MaxCompute that you can use for prediction or generation tasks. Models seamlessly integrate Artificial Intelligence (AI) computing capabilities, such as large language models and machine learning models, into the same platform where your data resides.
Model version: A model version is an independent and uniquely identifiable sub-object of a model. You can create and manage multiple iterative versions under the same model name. This simplifies grayscale releases, rapid rollbacks, and performance comparisons between old and new versions when you call models and AI Functions.
Benefits
Unified management: MaxCompute provides multiple model types. Similar to data, models support permission management and versioning. This helps you meet your enterprise's security and compliance requirements.
Multi-engine integration: MaxCompute supports model calls from various ecosystems, such as SQL and Python (MaxFrame). The unified architecture allows data analysts to use familiar SQL to call powerful AI models. It also allows data scientists to use the distributed Python computing capabilities of MaxFrame. They can combine these capabilities with models to continuously improve the efficiency and quality of data pre-processing.
Simplified Operations and Maintenance (O&M): You do not need to export data to external systems for AI inference. This avoids the security risks, costs, and latency issues associated with data movement.
Model types
MaxCompute provides different types of models:
Model type | Description | Tutorial |
Public model |
Note
| Use a MaxCompute public model for sentiment analysis of online reviews |
Remote model | You can connect to models that are already deployed on PAI-EAS. Specify the Endpoint and token required to access PAI-EAS to register the model as a MaxCompute remote model. Then, you can call the model using an AI Function. | Use a MaxCompute remote model to automatically generate E-commerce product descriptions |
Internally trained model | You can use MaxCompute MaxFrame to train traditional machine learning models. Execute | |
Import Model Imported model | In real-world business scenarios, built-in public models may not fully meet your needs. Models that are fine-tuned with algorithms can be adjusted based on business performance to achieve better results. You can import custom model files that are saved after being trained and tuned externally. Specify the OSS address of the model file to import it into MaxCompute for subsequent inference. | Being rolled out |
You can use the AI Functions provided by MaxCompute to call built-in public models or other types of models that you have created and managed in your project.
Model management
Before you manage models, make sure that your account has the permissions to manage model objects.
You can manage model objects in the following ways:
Management method
Instructions
You can manage models using SQL statements. This includes creating, viewing, modifying, and deleting models.
You can manage models using the MaxFrame Python language. Currently, only model creation is supported.
Manage models using the console
MaxCompute provides a graphical user interface (GUI) in the console for model management. In regions where the console has been adapted, you can view created models through the console.
Perform the following steps:
Log on to the MaxCompute console and select a region in the top-left corner.
In the navigation pane on the left, choose .
On the Projects page, find the target project and click Manage in its Actions column.
On the Project Settings page, click the Models tab.
You can view the public models and their version information in the
BIGDATA_PUBLIC_MODELSETpublic project. You can also view other types of models that you have created and their version information.
NoteThe feature to manage models in the console is currently available only in the China (Beijing) region. Support for other regions is being rolled out.