kempnerpulse.present.widgets

Rich panels and the line-plot renderable.

Every widget consumes ComputedRecord objects and reads metric values from the wrapped CanonicalRecord (canonical fractions/SI), converting to display units at render time via format. The responsive layout — fixed-width bars, stacked-vs-two-column card detail, fixed-width health/status badges, no-wrap headers, summary/footer field-drop ordering, fleet vertical scrolling, and the focus-view reflowing info grid — is reproduced from the single-file implementation, with the magic numbers kept as named constants.

Functions

build_fleet_panel(records, history, ...[, ...])	Lay out the fleet for an available width×height.
candidate_cols(n)	Sensible column counts for n cards: full row (n), full column (1), exact divisors, and balanced ragged grids; pruned so every row but the last is full.
choose_grid(n, W, H[, w_min])	Choose (cols, rows) for n cards in a W×H area, matching the grid aspect ratio to the window's.
fleet_panel(records, history[, ...])
gpu_card(rec, history[, power_limit, ...])
jobs_view_panel(records, gpu_processes)	Render a table of all running GPU compute processes with per-GPU metrics.
line_plot_view_panel(records, history[, ...])	Build the full Plot View: shared legend + 3×3 grid of line charts.
selected_gpu_panel(rec, history[, ...])
summary_panel(records, *[, app_version, ...])
workflow_label(weights)

Classes

GpuProcess	A single compute process running on a GPU (for the Job View).
LinePlotRenderable	Rich renderable that draws a line chart, adapting to available width.

kempnerpulse.present.widgets.workflow_label(weights)

Parameters:

weights (Tuple[float, float, float, float])

Return type:

str

class kempnerpulse.present.widgets.GpuProcess

Bases: object

A single compute process running on a GPU (for the Job View).

pid: int

user: str

gid: str

gpu_id: str

gpu_mem_mib: float | None

command: str

__init__(pid, user, gid, gpu_id, gpu_mem_mib, command)

Parameters:

• pid (int)

• user (str)

• gid (str)

• gpu_id (str)

• gpu_mem_mib (float | None)

• command (str)

Return type:

None

kempnerpulse.present.widgets.summary_panel(records, *, app_version='', cpu_info=(None, None, None, None), ram_info=(None, None), console_width=200)

Parameters:

• records (List[ComputedRecord])

• app_version (str)

• cpu_info (Tuple[int | None, int | None, float | None, int | None])

• ram_info (Tuple[float | None, float | None])

• console_width (int)

Return type:

Panel

kempnerpulse.present.widgets.gpu_card(rec, history, power_limit=None, nvlink_limit=None, detail_columns=2)

Parameters:

• rec (ComputedRecord)

• history (HistoryStore)

• power_limit (float | None)

• nvlink_limit (float | None)

• detail_columns (int)

Return type:

Panel

kempnerpulse.present.widgets.candidate_cols(n)

Sensible column counts for n cards: full row (n), full column (1), exact divisors, and balanced ragged grids; pruned so every row but the last is full.

Parameters:

n (int)

Return type:

List[int]

kempnerpulse.present.widgets.choose_grid(n, W, H, w_min=None)

Choose (cols, rows) for n cards in a W×H area, matching the grid aspect ratio to the window’s. Never more columns than fit at the card minimum width. Deterministic.

Parameters:

• n (int)

• W (int)

• H (int)

• w_min (int | None)

Return type:

Tuple[int, int]

kempnerpulse.present.widgets.fleet_panel(records, history, cards_per_row=2, detail_columns=2, power_limits=None, nvlink_bw_limits=None, avail_height=None, controller=None)

Parameters:

• records (List[ComputedRecord])

• history (HistoryStore)

• cards_per_row (int)

• detail_columns (int)

• power_limits (Dict[str, float] | None)

• nvlink_bw_limits (Dict[str, float] | None)

• avail_height (int | None)

• controller (CommandController | None)

Return type:

Panel

kempnerpulse.present.widgets.build_fleet_panel(records, history, avail_width, avail_height, power_limits=None, nvlink_bw_limits=None, controller=None)

Lay out the fleet for an available width×height. Shared by the main fleet view and the focus-mode mini-fleet so both behave identically.

Parameters:

• records (List[ComputedRecord])

• history (HistoryStore)

• avail_width (int)

• avail_height (int)

• power_limits (Dict[str, float] | None)

• nvlink_bw_limits (Dict[str, float] | None)

• controller (CommandController | None)

Return type:

Panel

class kempnerpulse.present.widgets.LinePlotRenderable

Bases: object

Rich renderable that draws a line chart, adapting to available width.

__init__(gpu_data, chart_rows=10, vmax=100.0, poll=1.0)

Parameters:

• gpu_data (List[Tuple[str, List[float]]])

• chart_rows (int)

• vmax (float)

• poll (float)

kempnerpulse.present.widgets.line_plot_view_panel(records, history, pcie_bw_limits=None, pcie_info='', poll=1.0, power_limits=None, console_height=50)

Build the full Plot View: shared legend + 3×3 grid of line charts.

Parameters:

• records (List[ComputedRecord])

• history (HistoryStore)

• pcie_bw_limits (Dict[str, float] | None)

• pcie_info (str)

• poll (float)

• power_limits (Dict[str, float] | None)

• console_height (int)

Return type:

Panel

kempnerpulse.present.widgets.selected_gpu_panel(rec, history, power_limit=None, nvlink_limit=None, console_width=200)

Parameters:

• rec (ComputedRecord)

• history (HistoryStore)

• power_limit (float | None)

• nvlink_limit (float | None)

• console_width (int)

Return type:

Panel

kempnerpulse.present.widgets.jobs_view_panel(records, gpu_processes)

Render a table of all running GPU compute processes with per-GPU metrics.

Parameters:

• records (List[ComputedRecord])

• gpu_processes (Dict[str, List[GpuProcess]])

Return type:

Panel