Line Plots

Log metric time series and interact with line charts in Pluto

Logging

Line plots are ideal for plotting numerical data over time — losses, accuracy, learning rates, and similar metrics.

Supported datatypes are float, int, or torch.Tensor with a single value, or arrays of such.

pluto.log({"train/train_loss": 2})

Viewing

Adding to a dashboard

Metrics appear automatically in the Charts view. To place a line chart on a custom dashboard, open Add Widget, choose the Metrics tab, and search for your metric.

Zooming

Pluto line charts support interactive zooming on both axes, letting you focus on specific regions of your data.

X-Axis Zoom

Click and drag horizontally on any chart to zoom into a range of steps. All charts in the same axis group (e.g., all step-based charts) will synchronize their zoom level, so zooming on one chart zooms all related charts to the same step range.

Y-Axis Zoom

Click and drag vertically on a chart to zoom the Y-axis. This is useful for focusing on a narrow value range, such as when losses are tightly clustered.

Y-axis zoom is independent per chart — it does not sync across panels.

Reset Zoom

Double-click on a chart to reset its zoom and return it to the full data range.

Fullscreen View

Click the expand button on any chart’s toolbar to open it in fullscreen.

The fullscreen view provides:

• A larger canvas for detailed inspection
• A resizable legend sidebar on the right showing all runs with color indicators
• Click any entry in the legend sidebar to toggle that series’ visibility — companion series (smoothing lines, min/max bands) toggle together automatically
• Export and Chart Settings buttons in the top-right corner
• The legend sidebar width is adjustable by dragging its edge, and persists across sessions

Tooltip

When hovering over a line chart, a tooltip appears showing the values for all visible runs at the nearest step. The tooltip can be pinned in place for hands-free inspection, customized to show exactly the columns you need, and searched to find specific series.

Hovering

Move your cursor over any chart to see a floating tooltip with the current step and values for each run. The tooltip follows your cursor and updates in real time as you move along the X-axis. A vertical crosshair line indicates the exact step position.

Pinning

Click anywhere on a chart to pin the tooltip at that step. A pinned tooltip:

• Stays fixed in place while you scroll or interact with other charts
• Has a blue border to distinguish it from the hover tooltip
• Can be dismissed by clicking the close button (x) in the top-right corner

Only one tooltip can be pinned at a time. Pinning a new tooltip replaces the previous one.

Dragging and Resizing

Once pinned, the tooltip becomes fully interactive:

• Drag the tooltip by its header to reposition it anywhere on screen
• Resize the tooltip by dragging the handle in the bottom-right corner

This is useful when the tooltip overlaps important chart data or when you want to place it next to another chart for comparison.

Searching Series

The pinned tooltip includes a search bar at the top. Type to filter the list of series by run name, display ID, or metric name. This is especially helpful when comparing many runs and you need to find a specific one.

Customizing Columns

The pinned tooltip shows one column per piece of run information. Four are on by default — Display ID, Run Name, Metric, and Value — and two more are opt-in: Min and Max.

To change them, click the gear icon at the top-right of a pinned tooltip (just left of the close button) to open the Tooltip Columns popover, then tick the columns you want. The gear only appears once the tooltip is pinned — a hover tooltip doesn’t have one.

Column choices and their order are saved to your browser and persist across sessions.

Raw Value. When smoothing is on, a Raw Value column is added automatically: Value shows the smoothed number, and Raw Value shows the original un-smoothed number at the same point. It appears only while smoothing is on, and isn’t listed in the popover.

A cell shows — when a value isn’t available for that series at that point. Interpolated values are prefixed with ~.

Chart Settings

Pluto provides two levels of chart configuration:

• Per-chart toolbar — Hover over any chart to reveal three buttons in the top-right corner: an export menu (copy to clipboard or download PNG), a chart settings popover (log scale toggles), and an expand button (fullscreen view).

  

• Line Chart Settings drawer — Accessible via the settings icon in the main toolbar. Configure global settings like X-axis metric, line width, max series, tooltip interpolation, and smoothing.

  

Export

Copy to Clipboard — Copy the current chart as an image directly to your clipboard for pasting into documents, Slack messages, or presentations.

Download as PNG — Download a high-resolution PNG of the chart to your local machine.

X Axis

Custom Metric — By default, charts use the step number as the X-axis. You can select any logged metric as the X-axis instead using the searchable combobox. This is available in both the line settings drawer and the dashboard widget configuration.

Logarithmic Scale — Toggle logarithmic scaling on the X-axis.

Y Axis

Logarithmic Scale — Toggle logarithmic scaling on the Y-axis.

Outlier Detection — Toggle automatic outlier detection on the Y-axis. When enabled, extreme outlier values are excluded from the Y-axis range calculation, preventing a single spike from compressing the rest of your data into a narrow band.

Line Display

Line Width — Adjust the line width using a slider that ranges from 0.5px to 5px (default: 1.5px). This setting is persisted globally across all charts.

Skip Missing Values — Toggle this option to skip gaps in data instead of interpolating between them. When enabled, the chart will not draw lines connecting data points across missing values, preventing misleading visual connections across data gaps.

Sampling

Controls how each bucket’s representative point is chosen when the server downsamples raw metrics for display.

• Average (with min/max envelope) (default) — Each bucket becomes the mean of the points inside it, drawn with a shaded min/max band so the extremes remain visible. Best for smooth trends.
• LTTB (preserve spikes) — Largest Triangle Three Buckets picks real data points that preserve the curve’s visual shape, so a one-step spike survives downsampling instead of being averaged away. Best when sharp spikes matter — gradient norms, learning-rate schedules.

The difference is easiest to see on a spiky metric. Both charts below plot the same train/grad_norm over 6,000 steps:

Under Average (left) each spike is averaged into its bucket’s mean, so the plotted line stays flat near 1.0 and the spikes survive only as the shaded band. Under LTTB (right) the spikes are real points on the line.

Sampling changes the Value a bucket reports — a mean under Average, an actual logged point under LTTB. The min/max band and the Min/Max tooltip columns are available either way.

Interpolation

When comparing runs that log at different step intervals, the tooltip may need to estimate values between data points. You can control this behavior in the line chart settings drawer:

• None — Only shows values at exact data points
• Linear — Interpolates linearly between neighboring data points
• Last known — Shows the most recent recorded value before the cursor position

Performance

Resolution — How many points (buckets) the server returns per series.

Option	Points per series
Auto (screen-fit) (default)	Fits roughly one point per 4 pixels of chart width, clamped to 200–3,000
High (500 points)	500
Max (1,000 points)	1,000
Ultra (3,000 points)	3,000

Increase it for sharper detail when you’re zoomed in or viewing a chart fullscreen. Higher resolutions transfer more data and take longer to render. Zooming respects this setting: a zoomed query asks for one point per visible step, capped at your chosen preset.

Include NaN/Inf-only metrics — Off by default. Metrics whose values are entirely NaN or Infinity are normally hidden, because Pluto’s fast metric-discovery path skips them. Turn this on to surface them — discovery then falls back to a slower query against the raw data, so leave it off unless you need it. Metrics found this way are flagged with a rose warning icon in the metric picker, tooltipped All values are NaN or Infinity in the selected run(s).

The side-by-side view has its own NaN/Inf toggle — the leftmost icon in its toolbar. It’s independent of this setting and resets when you leave the view.

NaN and Infinity markers — A non-finite value can’t be plotted on a line, so the chart marks it with a symbol instead, drawn in that run’s color:

Symbol	Value	Drawn
⊗	NaN	Near the bottom of the chart
△	+Infinity	Near the top of the chart
▽	−Infinity	Near the bottom of the chart

Markers are pinned to the chart edges rather than to the curve, so a run that diverges doesn’t drag the Y-axis or bury the trends you’re comparing.

Max Series per Chart — Set the maximum number of series a single chart will render. When a chart exceeds this limit, a warning is displayed instead of rendering the chart. The default is 500 series. Select No limit to disable the cap entirely (may be slow).

Step Deduplication

When multiple values are logged at the same step (e.g., from distributed training ranks or resumed runs), this setting uses the last-logged value per step instead of averaging all values. This eliminates artificial min/max spread in confidence bands caused by duplicate logging.

Smoothing

Smoothing helps you see the underlying trend in noisy metrics by applying a mathematical filter to the raw data. For details on smoothing algorithms and controls, see Smoothing.