Multi-View

Create multiple independent dashboards within a single project. Each view saves its own filters, grouping, and chart layouts — without affecting the others.

Multi-View is built on the new dashboard architecture. How to upgrade: open the project you want to upgrade and click "Upgrade to New Dashboard" in the top-right corner. New projects created via the web or the new SDK automatically use the Multi-View dashboard — no manual upgrade needed.

Before You Upgrade

• ⚠️ Note:
  • The upgrade duration depends on the number of experiments and charts in the project — please be patient
  • Experiments in progress cannot be upgraded
  • The upgrade cannot be rolled back to the old dashboard — please confirm carefully before proceeding!

💥 Breaking Changes:

• After upgrading to the new dashboard, you must upgrade the SDK to v0.9.0+ to create experiments, resume experiments, and log metrics
• Projects not yet upgraded can still log metrics and create experiments with SDK versions earlier than v0.9.0; however, SDK v0.9.0+ cannot log metrics to or create experiments in projects on the old dashboard
• The old dashboard will no longer be maintained after October 1, 2026 — we recommend upgrading as soon as possible

• What is Multi-View?
• Default Views and Custom Views
• Personal Views and Public Views
• Managing Views
• Chart Layout Sync

What is Multi-View?

When tuning models, you might want to watch every wiggle of the loss curve, compare final metrics across a dozen experiments, and curate a clean set of result charts for a report — all at the same time. Cramming every chart into a single dashboard gets messy fast.

Multi-View solves this: one project, multiple dashboards. Create views for different analysis scenarios — each view remembers its own chart configuration, so switching views means switching perspectives in one click.

What Multi-View helps you do:

• Cover different analysis needs: create dedicated views such as "Experiment Monitoring", "Model Comparison", and "Result Reproduction"
• Switch analysis scenarios instantly: no need to repeatedly adjust filters and layouts — just switch views
• Avoid configuration conflicts: changes in the current view never disturb what you've already organized in other views
• Improve team collaboration: fewer conflicts when multiple people adjust dashboards, keeping everyone's analysis environment stable

Default Views and Custom Views

When creating a view, choose between two types:

• Default view: automatically renders charts for all logged metrics — ideal for quick validation and phases where you want a full picture of your data
• Custom view: starts from a blank canvas where you hand-pick and configure the metric charts you care about — ideal when you have too many metrics, or need to focus on core comparisons during hyperparameter tuning

Note

A newly created custom view is a blank canvas — charts need to be added manually.

Personal Views and Public Views

Views support two visibility scopes — choose based on personal use or team collaboration:

• Personal view: visible only to you — organize it freely to match your own workflow
• Public view: visible to all project collaborators — ideal as a shared team dashboard

Managing Views

A project can have multiple views, with convenient management operations:

• Reorder: drag to sort, keeping frequently used views within easy reach
• Rename: rename anytime so each view's purpose is clear at a glance
• Pin: pin your most important views to the front
• Duplicate: quickly create a copy of an existing view and fine-tune from there

Chart Layout Sync

Within the same view, the layouts of multi-experiment comparison charts and single-experiment charts are bidirectionally synced — any adjustment to charts or groups is mirrored between the two. "Edit once, apply everywhere", keeping the experience consistent across every dimension.

About Empty Charts

Because of layout sync, all experiments in a view share one unified chart layout. If experiment A logs train/loss but experiment B does not, B's single-experiment dashboard still keeps an empty train/loss chart to preserve the shared layout. The more the metrics differ between experiments, the more empty charts may appear — simply delete them manually.