Skip to main content

Using Model Evaluation

Evaluation Status​

Evaluation Details

In the "Status" column of the evaluation task list, you can view the current status of each evaluation task:

  • In Progress: The task has started running. At this stage, you cannot view details, but you can delete the task if needed.
  • Completed: The evaluation is finished. You can click "Detail" to view detailed metrics and comparison results.
  • Failed: The evaluation has failed; please inspect logs for specific error reasons.

Evaluation Details​

In the list of completed evaluation tasks, click the "Detail" button to view detailed evaluation results.

Evaluation Details
  • Task Metadata Overview: The top section displays the creation time, evaluated model list with specific commit hashes, allocated compute resources (e.g., A100), and framework runtime version (e.g., opencsg/evalscope:1.11.1-cu130).
  • Multi-Model / Multi-Version Comparison: If multiple models or versions were added during task creation, the results table provides side-by-side metric comparison (such as accuracy scores) across each model or commit, highlighting performance differences across model variants or fine-tuning checkpoints.
  • Per-Dataset Breakdown: Switch between the "Overview" and individual dataset tabs to inspect specific benchmark scores and metrics.

Download Results​

Click the "Download Results" button in the upper right corner of the details page (or in the task list) to download raw evaluation result artifacts and logs.

Delete Task​

In the list of completed evaluation tasks, click the "Delete" button to remove the evaluation task. This action cannot be undone, so please proceed with caution.