Skip to content

Commit 5470eb7

Browse files
committed
Smoke test fixes
1 parent 560711c commit 5470eb7

File tree

3 files changed

+10
-8
lines changed

3 files changed

+10
-8
lines changed

requirements-dev.txt

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -21,5 +21,5 @@ mkdocs>=1.5.0
2121
mkdocs-material>=9.2.0
2222

2323
# Development utilities
24-
ipython>=8.14.0
24+
ipython>=8.0.0
2525
jupyter>=1.0.0

requirements.txt

Lines changed: 3 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -30,5 +30,6 @@ pydantic>=2.0.0
3030
# Machine learning (for sampling)
3131
scikit-learn>=1.3.0
3232

33-
# Database drivers (PostgreSQL support)
34-
psycopg2-binary>=2.9.0
33+
# Database drivers (PostgreSQL support - optional)
34+
# Install separately if PostgreSQL support needed: pip install psycopg2-binary>=2.9.0
35+
# psycopg2-binary>=2.9.0

scripts/smoke_test.py

Lines changed: 6 additions & 5 deletions
Original file line numberDiff line numberDiff line change
@@ -214,17 +214,18 @@ async def run_minimal_benchmark(self) -> None:
214214
benchmark_name="Smoke Test Benchmark"
215215
)
216216

217-
if result.success:
217+
if result.is_successful:
218218
self.benchmark_id = result.benchmark_id
219219
console.print(f"[green]Benchmark completed successfully (ID: {result.benchmark_id})[/green]")
220-
console.print(f"[dim]Execution time: {result.execution_time:.2f}s[/dim]")
220+
console.print(f"[dim]Execution time: {result.execution_time_seconds:.2f}s[/dim]")
221221

222222
if result.metrics:
223-
console.print(f"[dim]Accuracy: {result.metrics.accuracy.overall_accuracy:.1%}[/dim]")
224-
console.print(f"[dim]Total cost: ${result.metrics.cost.total_cost:.4f}[/dim]")
223+
console.print(f"[dim]Accuracy: {result.accuracy_rate:.1%}[/dim]")
224+
console.print(f"[dim]Total cost: ${result.total_cost:.4f}[/dim]")
225225

226226
else:
227-
raise Exception(f"Benchmark failed: {result.error_message}")
227+
error_msg = "; ".join(result.errors) if result.errors else "Unknown error"
228+
raise Exception(f"Benchmark failed: {error_msg}")
228229

229230
except Exception as e:
230231
if "API" in str(e) or "OpenRouter" in str(e):

0 commit comments

Comments
 (0)