fix more errors

Author: Benny Chen

eval_protocol/benchmarks/test_livebench_data_analysis.py

@@ -258,6 +258,7 @@ def _read_jsonl_table_from_text(text: str, header_cols: List[str]):
 
     reader = _read_df_v1 if version == "v1" else _read_df_v2
     gt_df = reader(output_fmt, ground_truth)
+    assert gt_df is not None, "GT dataframe is None"
 
     llm_clean = _clean_llm_output(llm_answer)
     llm_clean = _remove_initial_phrase(llm_clean)

eval_protocol/benchmarks/test_tau_bench_airline.py

@@ -198,6 +198,7 @@ def test_tau_bench_airline_evaluation(row: EvaluationRow) -> EvaluationRow:
     task = Task(
         id="Filler", evaluation_criteria=evaluation_criteria, user_scenario=UserScenario(instructions="Filler")
     )  # id and user_scenario are required for the Task type but not used in calculating reward
+    assert task.evaluation_criteria is not None, "Task evaluation criteria is None"
 
     if RewardType.DB in task.evaluation_criteria.reward_basis:
         env_reward_info = EnvironmentEvaluator.calculate_reward(

eval_protocol/execution/pipeline.py

@@ -212,6 +212,7 @@ async def _execute_standard_generation(
         if system_prompt_content:
             current_messages_for_rollout.append({"role": "system", "content": system_prompt_content})
         current_messages_for_rollout.append({"role": "user", "content": user_query})
+        assert self.model_client is not None, "at this point model client needs to be initialized"
 
         generation_output_std = await self.model_client.generate(
             messages=current_messages_for_rollout,

eval_protocol/rewards/apps_testing_util.py

@@ -174,7 +174,10 @@ def run_test(in_outs, test=None, debug=False, timeout=15):
                 if isinstance(last_block, ast.If):
                     condition = last_block.test
                     if ast.unparse(condition).strip() == "__name__ == '__main__'":
-                        test = ast.unparse(astree.body[:-1]) + "\n" + ast.unparse(last_block.body)
+                        # Build modules for unparse to avoid passing lists to ast.unparse
+                        prefix_module = ast.Module(body=astree.body[:-1], type_ignores=[])
+                        body_module = ast.Module(body=last_block.body, type_ignores=[])
+                        test = ast.unparse(prefix_module) + "\n" + ast.unparse(body_module)
             except Exception:
                 pass