Fix Completion Params

eval_protocol/pytest/evaluation_test.py

@@ -54,6 +54,7 @@
     parse_ep_max_rows,
     parse_ep_max_concurrent_rollouts,
     parse_ep_num_runs,
+    parse_ep_completion_params,
     rollout_processor_with_retry,
     sanitize_filename,
 )
@@ -334,10 +335,14 @@ def evaluation_test(  # noqa: C901
 
     active_logger: DatasetLogger = logger if logger else default_logger
 
-    # Apply override from pytest flags if present
+    # Optional global overrides via environment for ad-hoc experimentation
+    # EP_INPUT_PARAMS_JSON can contain a JSON object that will be deep-merged
+    # into input_params (e.g., '{"temperature":0,"extra_body":{"reasoning":{"effort":"low"}}}').
     num_runs = parse_ep_num_runs(num_runs)
     max_concurrent_rollouts = parse_ep_max_concurrent_rollouts(max_concurrent_rollouts)
     max_dataset_rows = parse_ep_max_rows(max_dataset_rows)
+    completion_params = parse_ep_completion_params(completion_params)
+    original_completion_params = completion_params
 
     def decorator(
         test_func: TestFunction,
@@ -420,9 +425,6 @@ async def execute_with_params(
             else:
                 return test_func(**kwargs)
 
-        # preserve the original completion_params list for groupwise mode
-        original_completion_params_list = completion_params
-
         # Calculate all possible combinations of parameters
         if mode == "groupwise":
             combinations = generate_parameter_combinations(
@@ -544,20 +546,6 @@ def _log_eval_error(status: Status, rows: Optional[List[EvaluationRow]] | None,
                             "No model provided. Please provide a model in the completion parameters object."
                         )
 
-                    # Optional global overrides via environment for ad-hoc experimentation
-                    # EP_INPUT_PARAMS_JSON can contain a JSON object that will be deep-merged
-                    # into input_params (e.g., '{"temperature":0,"extra_body":{"reasoning":{"effort":"low"}}}').
-                    try:
-                        import json as _json
-
-                        _env_override = os.getenv("EP_INPUT_PARAMS_JSON")
-                        if _env_override:
-                            override_obj = _json.loads(_env_override)
-                            if isinstance(override_obj, dict):
-                                completion_params = deep_update_dict(dict(completion_params), override_obj)
-                    except Exception:
-                        pass
-
                     # Create eval metadata with test function info and current commit hash
                     eval_metadata = EvalMetadata(
                         name=test_func.__name__,
@@ -661,7 +649,7 @@ async def _execute_eval_with_semaphore(**inner_kwargs):
                             row_groups = defaultdict(list)  # key: row_id, value: list of rollout_result
                             tasks: List[asyncio.Task[List[EvaluationRow]]] = []
                             # completion_groups = []
-                            for idx, cp in enumerate(original_completion_params_list):
+                            for idx, cp in enumerate(original_completion_params):
                                 config = RolloutProcessorConfig(
                                     completion_params=cp,
                                     mcp_config_path=mcp_config_path or "",
@@ -744,7 +732,7 @@ async def _collect_result(config, lst):
                     # rollout_id is used to differentiate the result from different completion_params
                     if mode == "groupwise":
                         results_by_group = [
-                            [[] for _ in range(num_runs)] for _ in range(len(original_completion_params_list))
+                            [[] for _ in range(num_runs)] for _ in range(len(original_completion_params))
                         ]
                         for i_run, result in enumerate(all_results):
                             for r in result:
@@ -757,7 +745,7 @@ async def _collect_result(config, lst):
                                 threshold,
                                 active_logger,
                                 mode,
-                                original_completion_params_list[rollout_id],
+                                original_completion_params[rollout_id],
                                 test_func.__name__,
                                 num_runs,
                             )

eval_protocol/pytest/utils.py

@@ -170,6 +170,25 @@ def parse_ep_max_concurrent_rollouts(default_value: int) -> int:
     return int(raw) if raw is not None else default_value
 
 
+def parse_ep_completion_params(completion_params: List[CompletionParams]) -> List[CompletionParams]:
+    """Apply EP_INPUT_PARAMS_JSON overrides to completion_params.
+
+    Reads the environment variable set by plugin.py and applies deep merge to each completion param.
+    """
+    try:
+        import json as _json
+
+        _env_override = os.getenv("EP_INPUT_PARAMS_JSON")
+        if _env_override:
+            override_obj = _json.loads(_env_override)
+            if isinstance(override_obj, dict):
+                # Apply override to each completion_params item
+                return [deep_update_dict(dict(cp), override_obj) for cp in completion_params]
+    except Exception:
+        pass
+    return completion_params
+
+
 def deep_update_dict(base: dict, override: dict) -> dict:
     """Recursively update nested dictionaries in-place and return base."""
     for key, value in override.items():