Change defaults of tau benchmarks and unpack extra_body more explicitly

Author: xzrderek

eval_protocol/benchmarks/test_tau_bench_airline.py

@@ -119,7 +119,7 @@ def tau_bench_airline_to_evaluation_row(data: List[Dict[str, Any]]) -> List[Eval
     rollout_processor=MCPGymRolloutProcessor(),
     rollout_processor_kwargs={"domain": "airline"},
     passed_threshold={"success": 0.4, "standard_error": 0.02},
-    num_runs=8,
+    num_runs=4,
     mode="pointwise",
     max_concurrent_rollouts=50,
     server_script_path=_get_server_script_path(),

eval_protocol/benchmarks/test_tau_bench_retail.py

@@ -109,7 +109,7 @@ def tau_bench_retail_to_evaluation_row(data: List[Dict[str, Any]]) -> List[Evalu
     rollout_processor=MCPGymRolloutProcessor(),
     rollout_processor_kwargs={"domain": "retail"},
     passed_threshold={"success": 0.65, "standard_error": 0.02},
-    num_runs=8,
+    num_runs=4,
     mode="pointwise",
     max_concurrent_rollouts=50,
     server_script_path=get_server_script_path(),

eval_protocol/pytest/default_mcp_gym_rollout_processor.py

@@ -223,7 +223,7 @@ def __call__(self, rows: List[EvaluationRow], config: RolloutProcessorConfig) ->
                     model_id=config.completion_params.get("model", None),
                     temperature=config.completion_params.get("temperature", 0.0),
                     max_tokens=config.completion_params.get("max_tokens", 4096),
-                    reasoning_effort=config.completion_params.get("reasoning_effort", None),
+                    **(config.completion_params.get("extra_body", {}) or {}),
                 )
 
             except Exception as e: