DataWorks Data Map provides topology visualization and full lifecycle management for AI assets through a unified metadata hub, improving asset governance efficiency. This topic describes how to view AI assets in DataWorks Data Map.
How it works
Data Map integrates PAI (Platform for AI) asset metadata so you can search, discover, and manage four AI asset types from a single interface: datasets, AI models, algorithm tasks, and model services.
For each asset type, lineage is generated automatically when you perform standard operations — such as registering a dataset, training a model, or deploying a service. The exception is distributed training tasks submitted in Deep Learning Containers (DLC), which require manual lineage reporting via SDK.
The lineage graph shows aggregated version lineage, meaning you can view lineage across all versions of an AI asset.
Use cases
AI asset governance: Trace where assets come from and how they are used to improve data and model quality and meet compliance requirements.
Model traceability: In the context of Responsible AI, track the datasets, feature engineering methods, and parameter tuning processes behind each model — essential for regulatory audits and validating experiment results.
Troubleshooting: When model prediction accuracy drops, use the lineage graph to trace changes in upstream data processing and quickly identify the root cause.
Resource optimization: Understand task dependencies to identify which tasks can run in parallel, avoid redundant computations, and reduce costs.
Cross-team collaboration: Share clear lineage information across teams that use the same infrastructure, accelerating research and innovation.
Prerequisites
Before you begin, ensure that you have:
A workspace for the new version of Data Studio, with one of the following editions:
Standard Edition: Required for standard AI asset lineage services
Professional Edition: Required for lineage reporting on DLC model training tasks and workflow tasks
If you created a PAI workspace, select Create AI Workspace with Same Name during workspace creation. This enables Data Map to search and view the corresponding AI asset types.
Supported AI assets
Data Map lets you search for and view metadata details for the following AI assets:
Limitations
| Limitation | Details |
|---|---|
| DLC distributed training tasks | Lineage must be manually reported. Import the PAI SDK in your task code and configure inputs and outputs based on your task conditions. Incorrect configuration affects lineage accuracy. Requires whitelist access — contact your business manager. See Model training task. |
| Edition requirement | Standard Edition is required for standard AI asset lineage. Professional Edition is required for DLC model training tasks and workflow task lineage reporting. |
Collect AI assets
After you enable the PAI service for your workspace, Data Map automatically collects asset information from your AI workspace.
Search for AI assets
Log on to the DataWorks console. In the top navigation bar, select the target region. In the left-side navigation pane, choose Data Governance > Data Map, then click Go to Data Map.
Click the
icon in the left navigation pane to go to the Search page.In the Type section on the left side, select the AI asset type you want to view, then enter keywords in the search box.
View AI asset lineage
The Search page in Data Map supports viewing lineage information for all AI asset types. Each asset's details page also has a direct entry point for lineage analysis.
Dataset lineage
How lineage is generated
Lineage is generated automatically when you perform any of the following operations.
Method 1: Register a storage path as a dataset
When you register a storage path as a dataset, the system automatically reports lineage based on the data source:
| Data source | Lineage generated |
|---|---|
| OSS | From OSS to the dataset |
| General-purpose NAS file system or Extreme NAS file system | From NAS to the dataset |
| CPFS or CPFS for LINGJUN | From CPFS to the dataset |
Method 2: Register a data table as a dataset
When the data source is a MaxCompute table, lineage is generated from the MaxCompute table to the dataset:
If multiple SQL tasks generated the MaxCompute table, the dataset's lineage is traced back to all related SQL tasks.
Method 3: Annotate a dataset in iTAG
After a data annotation task is successfully created in iTAG, the system automatically reports the following lineage:
View dataset lineage
From Data Map
Log on to the DataWorks console. In the top navigation bar, select the target region. In the left-side navigation pane, choose Data Governance > Data Map, then click Go to Data Map.
Click the
icon in the left navigation pane to go to the Search page.In the Type section, select Dataset. In the Filter Conditions section, filter by Workspace and Data Source Type to narrow the list.
Find the dataset and click View Lineage.
On the lineage details page:
Lineage Time Range (area ①): Select a time range to view the dataset's lineage for the corresponding period.
Dataset Version (area ②): Switch between versions to view version-specific lineage.
Right-click the dataset node to select View Details or View Lineage.
From PAI
Log on to the PAI console. Switch to the target region, select Workspace in the left navigation pane, and click the target workspace.
In the left navigation pane, choose AI Asset Management > Datasets.
On the Custom Dataset tab, find and click the dataset name.
Click View Lineage to open the Lineage Analysis page for the current dataset version.
On the PAI lineage analysis page, right-click any node and select View Lineage to go to Data Map and view upstream and downstream lineage for that node.
AI model lineage
How lineage is generated
Lineage is generated automatically when you perform any of the following operations.
Method 1: Execute a model training task
After a model training task submitted in Model Gallery completes successfully, lineage is generated between the model and the training task:
Method 2: Register a model
After you register a model on the Model Management page, lineage is generated from the model storage path to the model:
Method 3: Deploy a model
After you deploy a registered model to an EAS service, lineage is generated from the model to the model service:
View AI model lineage
From Data Map
Log on to the DataWorks console. In the top navigation bar, select the target region. In the left-side navigation pane, choose Data Governance > Data Map, then click Go to Data Map.
Click the
icon in the left navigation pane to go to the Search page.In the Type section, select AI Model. In the Filter Conditions section, filter by Workspace.
Find the AI model and click View Lineage.
On the lineage details page:
Lineage Time Range (area ①): Select a time range to view the AI model's lineage for the corresponding period.
AI Model Version (area ②): Switch between versions to view version-specific lineage.
Right-click the AI model node to select View Details or View Lineage.
From PAI
Log on to the PAI console. Switch to the target region, select Workspace in the left navigation pane, and click the target workspace.
In the left navigation pane, choose AI Asset Management > Models.
On the Model page, find your registered model and click the latest version number to open the Model Version page.
In the Version Overview section, click View Lineage to open the Lineage Analysis page.
On the PAI lineage analysis page, right-click any node and select View Lineage to go to Data Map and view upstream and downstream lineage for that node.
Algorithm task lineage
How lineage is generated
Lineage is generated automatically when you perform any of the following operations.
For tasks submitted in DLC, you must manually report lineage by importing the SDK and configuring inputs and outputs. See Limitations for details.
Method 1: Execute a training task (algorithm component task)
After a model training task submitted in Model Gallery completes successfully, lineage is generated between the model and the training task:
Method 2: Execute a PAIflow task
After a PAIFlow task completes successfully, lineage is generated between the input table or storage path, the output model or output dataset, and the PAIFlow task:
Lineage is automatically collected and reported for PAIFlow tasks executed in the following ways:
Tasks submitted via the PAIFlow OpenAPI. For details, see API overview.
Tasks manually executed in Designer. For details, see Build and debug a model.
Tasks periodically triggered by DataWorks offline scheduling. For details, see Use DataWorks to schedule Designer pipelines offline.
Method 3: Execute a distributed training task
When you create and run a distributed training task, lineage is generated between the input table storage path, the output model or output dataset, and the DLC task:
For DLC tasks, lineage must be manually reported. See Limitations.
View algorithm task lineage
From Data Map
Log on to the DataWorks console. In the top navigation bar, select the target region. In the left-side navigation pane, choose Data Governance > Data Map, then click Go to Data Map.
Click the
icon in the left navigation pane to go to the Search page.In the Type section, select Algorithm Task. In the Filter Conditions section, filter by Workspace.
Find the algorithm task and click View Lineage.
On the lineage details page:
Lineage Time Range: Select a time range to view the algorithm task's lineage for the corresponding period.
Right-click the algorithm task node to select View Details or View Lineage.
From PAI
Log on to the PAI console. Switch to the target region, select Workspace in the left navigation pane, and click the target workspace.
In the left navigation pane, choose AI Asset Management > Jobs.
Find your task on one of the following tabs: Deep Learning Containers (DLC), Algorithm Component Job, or Pipeline Tasks.
Click the task name to open the task details page.
In the Basic Information section, click View Lineage to open the Lineage Analysis page.
On the PAI lineage analysis page, right-click any node and select View Lineage to go to Data Map and view upstream and downstream lineage for that node.
Model service lineage
How lineage is generated
After you deploy a registered model to an EAS service, the system automatically establishes lineage between the model and the model service:
View model service lineage
From Data Map
Log on to the DataWorks console. In the top navigation bar, select the target region. In the left-side navigation pane, choose Data Governance > Data Map, then click Go to Data Map.
Click the
icon in the left navigation pane to go to the Search page.In the Type section, select Model Service. In the Filter Conditions section, filter by Status and Resource Group Name.
Find the model service and click View Lineage.
On the lineage details page:
Lineage Time Range: Select a time range to view the model service's lineage for the corresponding period.
Right-click the model service node to select View Details or View Lineage.
From PAI
Log on to the PAI console. Switch to the target region, select Workspace in the left navigation pane, and click the target workspace.
In the left navigation pane, choose Model Deployment > Elastic Algorithm Service (EAS).
On the Inference Service tab, find your model service and click the task name.
In the Basic Information section, click View Lineage to open the Lineage Analysis page.
On the PAI lineage analysis page, right-click any node and select View Lineage to go to Data Map and view upstream and downstream lineage for that node.