mirror of
https://github.com/elder-plinius/OBLITERATUS.git
synced 2026-08-17 16:37:30 +02:00
109 lines
3.2 KiB
Python
109 lines
3.2 KiB
Python
"""CPU-only contracts for benchmark visualization generation."""
|
|
|
|
from __future__ import annotations
|
|
|
|
import matplotlib.pyplot as plt
|
|
import pytest
|
|
|
|
from obliteratus.evaluation import benchmark_plots as plots
|
|
|
|
|
|
@pytest.fixture
|
|
def results():
|
|
return [
|
|
{
|
|
"method": "advanced",
|
|
"model": "org/model-a",
|
|
"model_short": "model-a",
|
|
"perplexity": 10.0,
|
|
"refusal_rate": 0.1,
|
|
"coherence": 0.9,
|
|
"time_s": 2.0,
|
|
"strong_layers": 3,
|
|
"ega_expert_dirs": 2,
|
|
"cot_preserved": 1,
|
|
"expert_classified_layers": 4,
|
|
},
|
|
{
|
|
"method": "unknown/hf_secret123456",
|
|
"model": "org/model-b",
|
|
"perplexity": 12.0,
|
|
"refusal_rate": 0.2,
|
|
"coherence": 0.8,
|
|
"time_s": 3.0,
|
|
"strong_layers": 0,
|
|
"ega_expert_dirs": 0,
|
|
"cot_preserved": 0,
|
|
"ega_safety_layers": 0,
|
|
},
|
|
]
|
|
|
|
|
|
def test_label_sanitization_and_palette_fallback():
|
|
label = plots._sanitize_label(
|
|
"/private/path/model hf_secret123456 0123456789abcdef0123456789abcdef", max_len=40,
|
|
)
|
|
assert "/private/path" not in label
|
|
assert "hf_secret" not in label
|
|
assert label == "model <TOKEN> <REDACTED>"
|
|
assert plots._get_color("basic") == plots.PALETTE["basic"]
|
|
assert plots._get_color("custom", 9) == plots.MODEL_PALETTE[1]
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
"factory",
|
|
[
|
|
plots.plot_pareto_frontier,
|
|
plots.plot_method_radar,
|
|
plots.plot_metric_bars,
|
|
plots.plot_timing_efficiency,
|
|
plots.plot_moe_metrics,
|
|
plots.plot_model_scaling,
|
|
],
|
|
)
|
|
def test_each_plot_returns_a_figure_for_empty_input(factory):
|
|
figure = factory([])
|
|
assert isinstance(figure, plt.Figure)
|
|
plt.close(figure)
|
|
|
|
|
|
@pytest.mark.parametrize(
|
|
"factory",
|
|
[
|
|
plots.plot_pareto_frontier,
|
|
plots.plot_method_radar,
|
|
plots.plot_metric_bars,
|
|
plots.plot_timing_efficiency,
|
|
plots.plot_moe_metrics,
|
|
plots.plot_model_scaling,
|
|
],
|
|
)
|
|
def test_each_plot_renders_synthetic_results(factory, results):
|
|
figure = factory(results, " — fixture")
|
|
assert isinstance(figure, plt.Figure)
|
|
assert figure.axes
|
|
plt.close(figure)
|
|
|
|
|
|
def test_pareto_handles_single_point_without_frontier_line(results):
|
|
figure = plots.plot_pareto_frontier(results[:1])
|
|
assert isinstance(figure, plt.Figure)
|
|
plt.close(figure)
|
|
|
|
|
|
def test_moe_plot_handles_non_moe_results(results):
|
|
figure = plots.plot_moe_metrics(results[1:])
|
|
assert "No MoE-specific features" in figure.axes[0].texts[0].get_text()
|
|
plt.close(figure)
|
|
|
|
|
|
def test_dashboard_modes_and_optional_moe_panel(results):
|
|
assert plots.generate_benchmark_dashboard([]) == []
|
|
method_figures = plots.generate_benchmark_dashboard(results, mode="multi_method")
|
|
model_figures = plots.generate_benchmark_dashboard(results, mode="multi_model")
|
|
assert len(method_figures) == 5
|
|
assert len(model_figures) == 5
|
|
assert plots.generate_benchmark_dashboard(results, mode="unsupported") == []
|
|
for figure in method_figures + model_figures:
|
|
plt.close(figure)
|