Tags are a core component of application evaluation and observability, supporting custom annotation of evaluation data and observability data. By creating different types of tags, you can build an evaluation dimension system tailored to your business scenarios and analyze application output quality from multiple perspectives.
Tags are a core component of application evaluation and observability, supporting custom annotation of evaluation data and observability data. By creating different types of tags, you can build an evaluation dimension system tailored to your business scenarios and analyze application output quality from multiple perspectives.
Go to the Tag Management page, click Create Tag to enter the tag creation page, and configure the following:
The system supports the following four tag types for different annotation scenarios:
A categorical tag selects one or more values from predefined options—suitable for enumerated evaluation dimensions.
Example use cases: Answer quality (poor / fair / good), error type (factual / logical / formatting), sentiment (positive / neutral / negative).
A boolean tag is for yes/no judgments—suitable for binary evaluation dimensions.
Example use cases: Is correct, contains hallucination, meets standards, is complete.
A numeric tag is for numeric scoring—suitable for evaluation dimensions requiring precise scores.
Example use cases: Score (1-5), relevance score (0-1), completeness percentage (0-100).
A text tag is for free-text input—suitable for evaluation dimensions requiring detailed descriptions or notes.
Example use cases: Error reason, improvement suggestion, annotation note, specific issue description.
Go to the Tag Management page to view and manage all created tags. Click Edit to modify a tag's configuration.
Go to the Evaluation Tasks page and click Details to the right of a task. You can annotate and filter evaluation data using tags.
Evaluation tasks support two annotation modes:
Click Annotate in the Actions column to view a record's full information:
On the Evaluation Tasks page, click Details to the right of a task, then select the Metrics page to view tag-based statistical analysis results.
In the observability module, you can annotate each Span with tags to assess the quality of real online data.
In the filter of the observability page, tags that a user has manually added automatically appear as filter options.
Create a Tag
Go to the Tag Management page, click Create Tag to enter the tag creation page, and configure the following:
| Configuration | Description |
|---|---|
| Tag name | Required, 1-50 characters. Used to identify and distinguish tags—use clear, understandable names. |
| Description | Optional, 0-200 characters. Describes the tag's purpose, criteria, or usage instructions. |
| Type | Choose the tag type. Categorical, Boolean, Numeric, and Text types are supported. See "Tag Types" below. |
Tag Types
The system supports the following four tag types for different annotation scenarios:
Categorical Tag
A categorical tag selects one or more values from predefined options—suitable for enumerated evaluation dimensions.
| Configuration | Description |
|---|---|
| Options | Custom options are supported, up to 20 options. |
| Annotation method | Select via a multi-select dropdown—one or more options can be chosen. |
| Filter conditions | Supported conditions: belongs to, does not belong to. |
Boolean Tag
A boolean tag is for yes/no judgments—suitable for binary evaluation dimensions.
| Configuration | Description |
|---|---|
| Options | Fixed at two options, True and False. |
| Annotation method | Select one of two options. |
| Filter conditions | Supported conditions: belongs to, does not belong to. |
Numeric Tag
A numeric tag is for numeric scoring—suitable for evaluation dimensions requiring precise scores.
| Configuration | Description |
|---|---|
| Data type | Supports Double numeric input—integers or decimals. |
| Annotation method | Enter a specific value via a numeric input box. |
| Filter conditions | Supported conditions: equals, not equals, greater than, greater than or equal, less than, less than or equal. |
Text Tag
A text tag is for free-text input—suitable for evaluation dimensions requiring detailed descriptions or notes.
| Configuration | Description |
|---|---|
| Data type | Supports String text input—any text content. |
| Annotation method | Enter specific content via a text input box. |
| Filter conditions | Supported conditions: contains, does not contain. |
Manage Tags
Go to the Tag Management page to view and manage all created tags. Click Edit to modify a tag's configuration.
Use Tags in Evaluation Tasks
Data Details
Go to the Evaluation Tasks page and click Details to the right of a task. You can annotate and filter evaluation data using tags.
Add Tags
- On the evaluation task details page, click Tag Configuration.
- In the side panel, select tags under the current workspace from the dropdown.
- Added tags appear in the evaluation task list, including tag name, tag type, and annotation status.
Annotation Modes
Evaluation tasks support two annotation modes:
| Mode | Description |
|---|---|
| Standard Mode | Fields are laid out horizontally. Click Annotate in the Actions column to view a record's full information and annotate one by one. |
| Quick Annotation | Custom tags become editable. Categorical and boolean tags appear as dropdowns; numeric and text tags appear as input boxes. Selections/inputs are saved immediately. |
Annotation View
Click Annotate in the Actions column to view a record's full information:
- Title: Shows "Annotation #ID," with previous/next navigation for continuous annotation.
- Dataset data: Displays dataset field names and content row by row, including input, output, and other fields.
- Application output: The application's chain-of-thought content (if any) and output data.
- Manual annotation: Displays all added tags; selections/changes are saved immediately.
Data Filtering
- Keyword search by filter type: Search field values for dataset content, application output, status, and result type via fuzzy matching.
- Filter by tag: Click Filter to add conditions, then click Apply to filter for specific content.
Metrics
On the Evaluation Tasks page, click Details to the right of a task, then select the Metrics page to view tag-based statistical analysis results.
Use Tags in Observability
In the observability module, you can annotate each Span with tags to assess the quality of real online data.
Add Annotations
- Click View Details to the right of a specific application, then click Data Annotation.
- In the side panel, select the tag to add. Click New Tag to jump to the Tag Management page to create a new tag.
- Annotations are saved automatically—no manual submission needed. After annotating, view them in the Tags column of the Span list.