Using Model Evaluation
After an evaluation task is created, you can track its progress, view results, and manage tasks from the evaluation task list.
Evaluation Status
Evaluation tasks go through the following statuses during execution:
| Status | Description |
|---|---|
| Evaluating | The task is running. You can open details to inspect the configuration snapshot and status; scores and coverage are written after success |
| Completed | The evaluation has finished; results and detailed reports are viewable |
| Failed | The evaluation task failed; check the failure reason and logs |
Note
View Evaluation Details
Click Details in the task list while the task is running or after it finishes.
The details page contains the following information:
| Information | Description |
|---|---|
| Overall Score | Composite score across all evaluation benchmarks (written after success) |
| Per-Benchmark Scores | Individual scores for each benchmark/dataset |
| Evaluation Metrics | Specific metrics such as accuracy, F1 score, BLEU score, etc. |
| Report | Configuration snapshot (framework, Prompt, scoring plugins, engine_args), sample coverage, and failed examples. See Evaluation Custom Parameters |
The evaluation details page gives you a comprehensive view of model performance across different benchmarks, providing data-driven insights for model selection and optimization.
Download Evaluation Results
On the evaluation details page or in the task list, click the Download button to download the complete evaluation results file.
The downloaded file contains:
- Evaluation score summary table
- Detailed evaluation data for each benchmark
- Sample model prediction outputs
Tip
Delete Evaluation Tasks
If an evaluation task is no longer needed, click the Delete button in the task list to remove it.
Warning