llm-eval-framework / tests /test_reporting.py
bhargob11's picture
Upload folder using huggingface_hub
69ee6e4 verified
Raw History Blame Contribute Delete
791 Bytes
from llm_eval.reporting import build_markdown_report
def test_markdown_report_contains_sections() -> None:
scored = {
"status": "completed",
"total_rows": 4,
"providers": {
"anthropic": {
"attempted": 2,
"correct": 2,
"accuracy": 1.0,
"avg_latency_ms": 123.0,
"errors": 0,
"accuracy_ci95": {"low": 0.34, "high": 1.0},
"categories": {"math": {"attempted": 2, "correct": 2, "accuracy": 1.0}},
}
},
}
pairwise = []
report = build_markdown_report("run123", scored, pairwise)
assert "# Evaluation Report: run123" in report
assert "## Leaderboard" in report
assert "## Category Breakdown" in report