nat.profiler.calc.plot

Attributes

logger

Classes

PlotConfig

Functions

plot_concurrency_vs_time_metrics_simple(→ None)	Save a simple plot of concurrency vs. p95 LLM latency and workflow runtime.
plot_metric_vs_concurrency_with_optional_fit(ax, x, y, ...)	Helper to plot a metric vs concurrency with pre-computed fit, outlier highlighting, and SLA line.
plot_concurrency_vs_time_metrics(→ None)	Plot concurrency vs. p95 latency and workflow runtime using metrics_per_concurrency.

Module Contents

logger

class PlotConfig

SIMPLE_FIGSIZE = (12, 6)

SIMPLE_LINEWIDTH = 2

SIMPLE_DPI = 150

ENHANCED_FIGSIZE = (16, 6)

ENHANCED_DPI = 300

DATA_MARKER = 'o'

OUTLIER_MARKER = 'x'

OUTLIER_COLOR = 'crimson'

TREND_COLOR = 'r'

TREND_LINESTYLE = '--'

TREND_ALPHA = 0.8

TREND_LINEWIDTH = 2.0

LLM_LATENCY_COLOR = 'steelblue'

RUNTIME_COLOR = 'darkgreen'

SLA_COLOR = 'red'

NOTE_BOX_COLOR = 'mistyrose'

NOTE_TEXT_COLOR = 'crimson'

STATS_BOX_COLOR = 'lightblue'

DATA_ALPHA = 0.7

OUTLIER_ALPHA = 0.9

GRID_ALPHA = 0.3

SLA_ALPHA = 0.7

NOTE_BOX_ALPHA = 0.7

STATS_BOX_ALPHA = 0.8

DATA_POINT_SIZE = 120

OUTLIER_POINT_SIZE = 140

DATA_LINEWIDTH = 1

AXIS_LABEL_FONTSIZE = 12

TITLE_FONTSIZE = 14

LEGEND_FONTSIZE = 10

NOTE_FONTSIZE = 10

STATS_FONTSIZE = 10

NOTE_X_POS = 0.98

NOTE_Y_POS = 0.02

STATS_X_POS = 0.02

STATS_Y_POS = 0.02

NOTE_BOX_PAD = 0.3

STATS_BOX_PAD = 0.5

TREND_LINE_POINTS = 100

AXIS_LABEL_FONTWEIGHT = 'bold'

TITLE_FONTWEIGHT = 'bold'

plot_concurrency_vs_time_metrics_simple(

df: pandas.DataFrame,

output_dir: pathlib.Path,

) → None

Save a simple plot of concurrency vs. p95 LLM latency and workflow runtime.

plot_metric_vs_concurrency_with_optional_fit(

ax: matplotlib.pyplot.Axes,

x: numpy.ndarray,

y: numpy.ndarray,

metric_name: str,

y_label: str,

title: str,

color: str,

sla_value: float = 0.0,

sla_label: str = None,

fit: nat.profiler.calc.data_models.LinearFitResult | None = None,

)

Helper to plot a metric vs concurrency with pre-computed fit, outlier highlighting, and SLA line. Requires pre-computed fit to be provided.

plot_concurrency_vs_time_metrics(

metrics_per_concurrency: dict[int, nat.profiler.calc.data_models.SizingMetrics],

output_dir: pathlib.Path,

target_llm_latency: float = 0.0,

target_runtime: float = 0.0,

llm_latency_fit: nat.profiler.calc.data_models.LinearFitResult | None = None,

runtime_fit: nat.profiler.calc.data_models.LinearFitResult | None = None,

) → None

Plot concurrency vs. p95 latency and workflow runtime using metrics_per_concurrency. Enhanced with better styling, trend analysis, and annotations. Only plots valid runs and requires pre-computed fits.