Access the Evaluation Template
After selecting the target FastGPT application for evaluation, a dedicated download button for the CSV evaluation template becomes available in the evaluation workspace. A visual reference for this interface element is provided via the asset at /imgs/evaluation2.png. This button generates a standardized CSV file pre-configured with the four mandatory fields required for evaluation dataset creation.
Required Dataset Fields and Constraints
The official FastGPT evaluation CSV template includes four core fields for each entry:
- Global variables: Contextual variables applied to the evaluation session
- q (question): The user input question submitted to the target application
- a (expected answer): The pre-defined correct response for comparison against the application’s output
- Chat history: Previous conversation context included during the evaluation query
All uploaded datasets must adhere to strict rules set by the platform:
- A maximum of 1,000 QA pairs per single dataset file
- Full compliance with the template’s field structure; unapproved additional fields will cause task initialization failures
- All data must be entered in strict alignment with the template’s format to ensure successful parsing during evaluation.
Step-by-Step Evaluation Task Creation
Follow these sequential steps to launch an evaluation task:
- Navigate to the evaluation section of your target FastGPT application.
- Select the specific application to run the evaluation against.
- Click the download CSV template button to retrieve the standardized dataset file.
- Populate the template with valid data across all required fields, ensuring adherence to the format constraints.
- Upload the fully completed CSV dataset file to the FastGPT evaluation interface.
- Click the "Start Evaluation" button to initialize the automated evaluation task.
Source: FastGPT official source
Applicability and version scope
Use this page for the documented Tutorials scenario. Confirm the FastGPT, dependency, API, and deployment versions in the official source before applying a change.
Safety guardrails
Use [REDACTED_CREDENTIAL] for credentials and private data. Confirm the documented environment and version before review.
Rollback guidance
Restore the prior technical-content authority snapshot. Restore saved configuration and data snapshots, then repeat the smallest verification scenario.