Dashboard & TableView
Dashboard and Table View are interfaces in Datumo Eval for understanding and analyzing evaluation results. The Dashboard enables quick understanding by visualizing overall performance trends, while Table View is configured as a detailed analysis interface for checking individual Query-level responses and evaluation reasoning. By using both interfaces together, you can create a natural analysis flow from overall performance to detailed cases.
Dashboard Concept
Dashboard provides summarized information to understand model performance at a high level. Through average scores, metric-specific performance, and visualization elements provided by evaluation type, you can quickly grasp overall model trends and patterns. It also provides functionality to compare multiple Eval Sets within a single Task, allowing you to view performance differences between models and version changes at a glance.
Table View Concept
Table View is a detailed analysis interface for examining individual evaluation items. For each Query, you can review model responses, scores, and Reasoning in order, and use filtering and search functions to intensively explore specific error types or areas of interest. Table View plays a critical role in situations requiring detailed debugging or outlier analysis.
Relationship Between Dashboard and Table View
While the two interfaces provide different perspectives, they are organically connected in the analysis process. Users can check overall performance trends in the Dashboard, then when they discover low performance in specific sections or abnormal patterns, they can analyze those items in detail in Table View. By repeating this flow, you can identify the root causes of model problems and derive improvement directions.
Leaderboard View
For standardized benchmark evaluations such as BEIR, results are provided in Leaderboard format. The Leaderboard is an interface configured to display rankings based on model scores, enabling objective comparison of multiple models.
Search, Filtering, and Export
Table View provides search based on Query, Response, and Metadata along with various filtering functions to quickly explore desired items. Dashboard also provides basic filtering functions for comparing multiple Eval Sets within a Task. Evaluation results can be downloaded in xlsx format for use in external analysis tools or report creation.