Better output in QA evaluations
This commit is contained in:
parent
98e0124ee0
commit
dae0992e3f
1 changed files with 5 additions and 15 deletions
|
|
@ -212,8 +212,6 @@ async def run_qa_benchmark(
|
||||||
return await qa.answer(question)
|
return await qa.answer(question)
|
||||||
|
|
||||||
for case in evaluation_dataset.cases:
|
for case in evaluation_dataset.cases:
|
||||||
progress.console.print(f"\n[bold]Evaluating case:[/bold] {case.name}")
|
|
||||||
|
|
||||||
single_case_dataset = EvalDataset[str, str, dict[str, str]](
|
single_case_dataset = EvalDataset[str, str, dict[str, str]](
|
||||||
cases=[case],
|
cases=[case],
|
||||||
evaluators=evaluation_dataset.evaluators,
|
evaluators=evaluation_dataset.evaluators,
|
||||||
|
|
@ -232,32 +230,24 @@ async def run_qa_benchmark(
|
||||||
result_case = report.cases[0]
|
result_case = report.cases[0]
|
||||||
|
|
||||||
equivalence = result_case.assertions.get("answer_equivalent")
|
equivalence = result_case.assertions.get("answer_equivalent")
|
||||||
progress.console.print(f"Question: {result_case.inputs}")
|
|
||||||
progress.console.print(f"Expected: {result_case.expected_output}")
|
|
||||||
progress.console.print(f"Generated: {result_case.output}")
|
|
||||||
if equivalence is not None:
|
if equivalence is not None:
|
||||||
progress.console.print(
|
|
||||||
f"Equivalent: {equivalence.value}"
|
|
||||||
+ (f" — {equivalence.reason}" if equivalence.reason else "")
|
|
||||||
)
|
|
||||||
if equivalence.value:
|
if equivalence.value:
|
||||||
passing_cases += 1
|
passing_cases += 1
|
||||||
|
|
||||||
progress.console.print("")
|
|
||||||
|
|
||||||
if report.failures:
|
if report.failures:
|
||||||
failures.extend(report.failures)
|
failures.extend(report.failures)
|
||||||
failure = report.failures[0]
|
failure = report.failures[0]
|
||||||
progress.console.print(
|
progress.console.print(
|
||||||
"[red]Failure encountered during case evaluation:[/red]"
|
"[red]Failure encountered during case evaluation:[/red]"
|
||||||
)
|
)
|
||||||
progress.console.print(f"Question: {failure.inputs}")
|
|
||||||
progress.console.print(f"Error: {failure.error_message}")
|
progress.console.print(f"Error: {failure.error_message}")
|
||||||
progress.console.print("")
|
progress.console.print("")
|
||||||
|
|
||||||
progress.console.print(
|
progress.update(
|
||||||
f"[green]Accuracy: {(passing_cases / total_processed):.4f} "
|
qa_task,
|
||||||
f"{passing_cases}/{total_processed}[/green]"
|
description="[yellow]Evaluating QA cases...[/yellow] "
|
||||||
|
f"[green]Accuracy: {(passing_cases / total_processed):.2f} "
|
||||||
|
f"{passing_cases}/{total_processed}[/green]",
|
||||||
)
|
)
|
||||||
progress.advance(qa_task)
|
progress.advance(qa_task)
|
||||||
|
|
||||||
|
|
|
||||||
Loading…
Reference in a new issue