Custom Dashboards

Custom dashboards let you build focused views of your experiment data. Instead of scrolling through every metric, you can curate sections with only the charts, images, and files you care about.

​

Overview

Dashboards are created at the project level and are shared across your organization. Each dashboard contains sections, and each section contains widgets. Widgets can display:

• Metrics — Line charts from numeric time-series data
• Files — Histograms, images, videos, audio, and console logs

​

Creating a Dashboard

1. Navigate to any project’s Compare tab
2. Click the Dashboards dropdown in the toolbar
3. Select Create New Dashboard
4. Enter a name for your dashboard and click Create

You can also import an existing dashboard configuration via the Import JSON tab. Once created, your dashboard appears in the Dashboards dropdown. Switch between the default Charts view and any custom dashboard at any time.

​

Editing a Dashboard

Click Edit Dashboard to enter edit mode. In edit mode you can:

• Add, remove, and reorder sections
• Add widgets to sections
• Choose between Grid and Free layout modes
• Save or Cancel your changes

​

Adding Sections

Click New Section to create a section. Each section has a name and acts as a collapsible container for widgets.

​

Adding Widgets

Inside a section, click Add Widget to open the widget type picker:

​

Metrics Widgets

Select Metrics to create a line chart widget. You can:

• Select multiple metrics to plot them on a single chart — each metric gets a distinct dash pattern for differentiation
• Search for metrics using fuzzy text search or glob patterns (e.g., train/*)
• Switch to Regex mode for precise pattern matching
• Configure the X-Axis (Step or Relative Time), Aggregation (Last, Min, Max, Mean), and Y-Axis Scale (Linear or Log)

When you select multiple metrics, each one is differentiated using distinct dash patterns (solid, dashed, dotted, etc.), making it easy to distinguish between them even in grayscale. This is useful for comparing known, related metrics (e.g., train/loss and val/loss on the same axis). Manually selected metrics are static — the chart always shows exactly the metrics you picked, regardless of what new data appears. For automatic discovery of new metrics, see Dynamic Content below.

​

File Widgets

Select Files to add widgets for non-numeric data:

• Histograms — Distribution visualizations with step navigation
• Images — Training visualizations, sample outputs, etc.
• Videos — Recorded training episodes or generated content
• Audio — Audio samples or generated speech
• Console Logs — Stdout/stderr from your runs with search and ANSI color support

​

Dynamic Content

Sections and widgets can be made dynamic, meaning they use a glob or regex pattern to automatically discover and display matching data. When new metrics or files appear that match the pattern, they are included automatically.

​

Dynamic Sections

Toggle Dynamic section when creating a section to auto-populate it with widgets from a pattern. A dynamic section creates one widget per match, automatically choosing the correct widget type for each. Dynamic sections are cross-data-type — a single pattern can match both metrics and files. For example, suppose you have the following logged data under the data/ prefix:

• data/loss (metric)
• data/accuracy (metric)
• data/sample_audio (audio file)
• data/confusion_matrix (histogram)
• data/predictions (video)

A dynamic section with the pattern data/* will create 5 separate widgets — one for each match — automatically choosing the correct widget type (line chart, audio player, histogram, video player) for each. As new data appears under the matched pattern, the section updates automatically. For example, if a new run logs data/f1_score, a 6th widget appears without any manual configuration.

​

Dynamic Widgets

When you use a glob pattern like train/* in a widget, the widget becomes dynamic — it automatically includes any matching data on a single widget. Unlike manually selecting multiple metrics (which is static), a dynamic widget will pick up new data as it appears. For example, suppose you have 5 metrics, train/loss1 through train/loss5, logged across your runs. A dynamic Metrics widget with the pattern train/* will plot all 5 metrics on one chart. If a new metric like train/loss6 is logged, it is automatically added to the same chart when the corresponding run is selected as visible. Because widgets are separated by data type, a dynamic Metrics widget only matches numeric time-series data. If you also have files under the same prefix, you would need a separate dynamic Files widget to display those.

​

Dynamic Sections vs Dynamic Widgets

Both use glob/regex patterns to automatically include matching data, but they differ in how matches are displayed: Example: You have train/loss, train/accuracy, train/sample_image, and train/audio_clip.

• A dynamic section with train/* creates 4 widgets — a line chart for loss, a line chart for accuracy, an image widget, and an audio widget
• A dynamic Metrics widget with train/* creates 1 chart with loss and accuracy plotted together (the image and audio are not shown — you’d need a separate dynamic Files widget for those)

​

Saving and Loading

Click Save to persist your dashboard. Dashboards are saved per-project and shared across all members of your organization. Anyone with access to the project can view and edit dashboards. Switch between dashboards using the Dashboards dropdown in the toolbar. Click Charts to return to the default view showing all metrics grouped automatically.