style(benchmarks): apply black formatting for truthfulqa PR

Author: saschabuehrle

cascadeflow/agent.py

@@ -1725,19 +1725,11 @@ async def _execute_direct_with_timing(
             prompt_tokens = response.metadata.get("prompt_tokens")
             completion_tokens = response.metadata.get("completion_tokens")
             total_tokens = response.metadata.get("total_tokens")
-            if (
-                total_tokens is None
-                and prompt_tokens is not None
-                and completion_tokens is not None
-            ):
+            if total_tokens is None and prompt_tokens is not None and completion_tokens is not None:
                 total_tokens = prompt_tokens + completion_tokens
 
         cost = None
-        if (
-            LITELLM_AVAILABLE
-            and prompt_tokens is not None
-            and completion_tokens is not None
-        ):
+        if LITELLM_AVAILABLE and prompt_tokens is not None and completion_tokens is not None:
             try:
                 provider = LiteLLMCostProvider()
                 cost = provider.calculate_cost(
@@ -2222,7 +2214,11 @@ def __init__(
                 self.speedup = 1.0
 
                 token_total = total_tokens
-                if token_total is None and prompt_tokens is not None and completion_tokens is not None:
+                if (
+                    token_total is None
+                    and prompt_tokens is not None
+                    and completion_tokens is not None
+                ):
                     token_total = prompt_tokens + completion_tokens
                 if token_total is None:
                     token_total = int(len(content.split()) * 1.3)

tests/benchmarks/base.py

@@ -429,7 +429,9 @@ def _generate_summary(self) -> BenchmarkSummary:
 
         # Cascade metrics
         direct_routed = sum(1 for r in valid_results if r.direct_routed)
-        drafter_accepted = sum(1 for r in valid_results if r.routing_strategy == "cascade" and r.accepted)
+        drafter_accepted = sum(
+            1 for r in valid_results if r.routing_strategy == "cascade" and r.accepted
+        )
         escalated = sum(1 for r in valid_results if r.verifier_rejected)
         cascade_total = drafter_accepted + escalated
 
@@ -447,14 +449,18 @@ def _generate_summary(self) -> BenchmarkSummary:
         p95_latency = latencies[p95_idx]
         cascadeflow_latencies = [r.cascadeflow_latency_ms for r in valid_results]
         avg_cascadeflow_latency = (
-            sum(cascadeflow_latencies) / len(cascadeflow_latencies) if cascadeflow_latencies else 0.0
+            sum(cascadeflow_latencies) / len(cascadeflow_latencies)
+            if cascadeflow_latencies
+            else 0.0
         )
 
         # Quality metrics
         correct = sum(1 for r in valid_results if r.is_correct)
         accuracy = (correct / len(valid_results) * 100) if valid_results else 0.0
 
-        drafter_results = [r for r in valid_results if r.routing_strategy == "cascade" and r.accepted]
+        drafter_results = [
+            r for r in valid_results if r.routing_strategy == "cascade" and r.accepted
+        ]
         drafter_correct = sum(1 for r in drafter_results if r.is_correct)
         drafter_accuracy = (
             (drafter_correct / len(drafter_results) * 100) if drafter_results else 0.0
@@ -468,9 +474,7 @@ def _generate_summary(self) -> BenchmarkSummary:
 
         direct_results = [r for r in valid_results if r.direct_routed]
         direct_correct = sum(1 for r in direct_results if r.is_correct)
-        direct_accuracy = (
-            (direct_correct / len(direct_results) * 100) if direct_results else 0.0
-        )
+        direct_accuracy = (direct_correct / len(direct_results) * 100) if direct_results else 0.0
 
         # Token usage
         total_input = sum(r.tokens_input for r in valid_results)
@@ -483,7 +487,9 @@ def _generate_summary(self) -> BenchmarkSummary:
             failed_tests=failed,
             drafter_accepted=drafter_accepted,
             escalated_to_verifier=escalated,
-            acceptance_rate_pct=(drafter_accepted / cascade_total * 100) if cascade_total > 0 else 0.0,
+            acceptance_rate_pct=(
+                (drafter_accepted / cascade_total * 100) if cascade_total > 0 else 0.0
+            ),
             escalation_rate_pct=(escalated / cascade_total * 100) if cascade_total > 0 else 0.0,
             direct_routed=direct_routed,
             direct_routing_pct=(direct_routed / successful * 100) if successful > 0 else 0.0,
@@ -524,9 +530,7 @@ def _print_summary(self, summary: BenchmarkSummary) -> None:
         print(
             f"  Escalated:           {summary.escalated_to_verifier} ({summary.escalation_rate_pct:.1f}%)"
         )
-        print(
-            f"  Direct Routed:       {summary.direct_routed} ({summary.direct_routing_pct:.1f}%)"
-        )
+        print(f"  Direct Routed:       {summary.direct_routed} ({summary.direct_routing_pct:.1f}%)")
 
         print("\nCOST ANALYSIS:")
         print(f"  Total Cost:          ${summary.total_cost:.6f}")

tests/benchmarks/bfcl/bfcl_full_benchmark.py

@@ -394,7 +394,9 @@ def _check_function_correct(
                 counts[func_key] = counts.get(func_key, 0) + 1
 
             for func, expected_count in counts.items():
-                func_mentions = response_lower.count(func) + response_lower.count(func.replace("_", " "))
+                func_mentions = response_lower.count(func) + response_lower.count(
+                    func.replace("_", " ")
+                )
                 if func_mentions < expected_count:
                     return False, True
             return True, True
@@ -404,9 +406,7 @@ def _check_function_correct(
         if expected_func is None:
             # No tool should be used
             func_correct = (
-                found_func is None
-                or "don't need" in response_lower
-                or "no tool" in response_lower
+                found_func is None or "don't need" in response_lower or "no tool" in response_lower
             )
             return func_correct, True
 

tests/benchmarks/metrics.py

@@ -145,9 +145,7 @@ def from_results(results: list[BenchmarkResult]) -> "QualityMetrics":
         direct_results = [r for r in results if r.direct_routed]
         direct_correct = sum(1 for r in direct_results if r.is_correct)
         direct_incorrect = len(direct_results) - direct_correct
-        direct_accuracy = (
-            (direct_correct / len(direct_results) * 100) if direct_results else 0.0
-        )
+        direct_accuracy = (direct_correct / len(direct_results) * 100) if direct_results else 0.0
 
         return QualityMetrics(
             overall_accuracy=overall_accuracy,

tests/benchmarks/mmlu/mmlu_full_benchmark.py

@@ -681,9 +681,7 @@ async def process_problem(index: int, problem: dict) -> tuple[int, BenchmarkResu
             )
             return index, benchmark_result, output
         except Exception as e:
-            output = (
-                f"[{index+1}/{len(problems)}] {problem['subject'][:20]:20s}: ⚠️ ERROR: {e}"
-            )
+            output = f"[{index+1}/{len(problems)}] {problem['subject'][:20]:20s}: ⚠️ ERROR: {e}"
             benchmark_result = BenchmarkResult(
                 query=problem["question"][:50] + "...",
                 subject=problem["subject"],
@@ -707,10 +705,7 @@ async def process_problem(index: int, problem: dict) -> tuple[int, BenchmarkResu
         for start in range(0, len(problems), concurrency):
             batch = problems[start : start + concurrency]
             batch_results = await asyncio.gather(
-                *(
-                    process_problem(start + offset, problem)
-                    for offset, problem in enumerate(batch)
-                )
+                *(process_problem(start + offset, problem) for offset, problem in enumerate(batch))
             )
             for index, benchmark_result, output in batch_results:
                 results[index] = benchmark_result

tests/benchmarks/tool_calls_realworld.py

@@ -480,11 +480,7 @@ async def run_cascade(self, query: Any) -> dict[str, Any]:
             "cascadeflow_latency_ms": (
                 (result.complexity_detection_ms or 0)
                 + (result.metadata.get("domain_detection_ms", 0) if result.metadata else 0)
-                + (
-                    result.metadata.get("tool_complexity_analysis_ms", 0)
-                    if result.metadata
-                    else 0
-                )
+                + (result.metadata.get("tool_complexity_analysis_ms", 0) if result.metadata else 0)
                 + (result.quality_verification_ms or 0)
             ),
             "tokens_input": result.metadata.get("prompt_tokens", 0),