fix smoke test params

Author: Dylan Huang

eval_protocol/pytest/evaluation_test.py

@@ -59,7 +59,7 @@ def evaluation_test(  # noqa: C901
     evaluation_test_kwargs: Optional[List[EvaluationInputParam]] = None,
     rollout_processor_kwargs: Optional[RolloutProcessorInputParam] = None,
     aggregation_method: AggregationMethod = "mean",
-    passed_threshold: Optional[Union[EvaluationThreshold, float]] = None,
+    passed_threshold: Optional[Union[EvaluationThreshold, float, dict]] = None,
     num_runs: int = 1,
     max_dataset_rows: Optional[int] = None,
     mcp_config_path: Optional[str] = None,

tests/test_tau_bench_airline_smoke.py

@@ -65,8 +65,13 @@ def tau_bench_airline_smoke_to_evaluation_row(data: List[Dict[str, Any]]) -> Lis
 @evaluation_test(
     input_dataset=["tests/pytest/data/airline_dataset.jsonl"],
     dataset_adapter=tau_bench_airline_smoke_to_evaluation_row,
-    model=["fireworks_ai/accounts/fireworks/models/gpt-oss-120b"],
-    rollout_input_params=[{"temperature": 0.8, "extra_body": {"reasoning_effort": "medium"}}],
+    completion_params=[
+        {
+            "temperature": 0.8,
+            "extra_body": {"reasoning_effort": "medium"},
+            "model": "fireworks_ai/accounts/fireworks/models/gpt-oss-120b",
+        }
+    ],
     rollout_processor=default_mcp_gym_rollout_processor,
     passed_threshold=0.36,
     num_runs=1,  # Smoke test: single run for quick feedback