fix: UTF-8 result writes, and make --force actually discard prior results

agent.py

@@ -682,7 +682,7 @@ def _handle_zep_agent(self, message, memorizing, query_id, context_id):
             # save the context
             save_dir = f"./outputs/rag_retrieved/{self.agent_name}/k_{self.retrieve_num}/{self.sub_dataset}/chunksize_{self.chunk_size}/query_{query_id}_context_{context_id}.json"
             os.makedirs(os.path.dirname(save_dir), exist_ok=True)
-            with open(save_dir, "w") as f:
+            with open(save_dir, "w", encoding="utf-8") as f:
                 paragraphs = [p for p in retrieved_context.replace("\r\n", "\n").split("\n") if p.strip()]
                 json.dump({"retrieved_context_paragraphs": paragraphs, "response": response}, f, ensure_ascii=False, indent=2)
 
@@ -746,8 +746,8 @@ def _process_rag_query(self, message, query_id, context_id):
         if output.get("retrieval_context"):
             save_dir = f"./outputs/rag_retrieved/{self.agent_name}/k_{self.retrieve_num}/{self.sub_dataset}/chunksize_{self.chunk_size}/query_{query_id}_context_{context_id}.json"
             os.makedirs(os.path.dirname(save_dir), exist_ok=True)
-            with open(save_dir, "w") as f:
-                json.dump(output["retrieval_context"], f)
+            with open(save_dir, "w", encoding="utf-8") as f:
+                json.dump(output["retrieval_context"], f, ensure_ascii=False)
 
             # drop the retrieval_context       
             output.pop("retrieval_context")
@@ -1091,13 +1091,13 @@ def save_agent(self):
             shutil.copyfile(source_db_path, target_db_path)
 
             # Save the agent ID for future loading
-            with open(f"{agent_save_folder}/agent_id.txt", "w") as f:
+            with open(f"{agent_save_folder}/agent_id.txt", "w", encoding="utf-8") as f:
                 f.write(self.agent_state.id)
         elif self._is_agent_type("zep"):
             # save the message that agent has processed
             messages = "agent finished memorization"
             os.makedirs(self.agent_save_to_folder, exist_ok=True)
-            with open(f"{self.agent_save_to_folder}/messages.txt", "w") as f:
+            with open(f"{self.agent_save_to_folder}/messages.txt", "w", encoding="utf-8") as f:
                 f.write(messages)
 
         print("\n\n Agent saved...\n\n")

initialization.py

@@ -60,23 +60,30 @@ def create_agent_and_fetch_data(agent_config, dataset_config):
     return start_time, conversation_creator.get_chunks(), conversation_creator.get_query_and_answers()
 
 
-def load_existing_results(output_file_path, dataset_config, all_query_answer_pairs):
+def load_existing_results(output_file_path, dataset_config, all_query_answer_pairs,
+                          force_rerun=False):
     """
     Load existing results from output file and initialize variables.
-    
+
     Args:
         output_file_path: Path to the output results file
         dataset_config: Configuration dictionary for the dataset
         all_query_answer_pairs: List of query-answer pairs for all contexts
-
+        force_rerun: When True, ignore any saved results and start from scratch.
+
     Returns:
         tuple: (metrics, results, last_completed_context_id, last_completed_query_id)
     """
-    if not os.path.exists(output_file_path):
+    # --force promises "re-run even if results already exist". Returning saved rows here breaks
+    # that promise twice: their metrics are averaged into the new score, and their high-water
+    # mark makes should_skip_query skip the very queries the flag asked to re-run. A run after a
+    # code change then reports figures partly produced by the old code, with nothing in the
+    # output to show it happened.
+    if force_rerun or not os.path.exists(output_file_path):
         return defaultdict(list), [], 0, 0
-    
+
     # Load existing results from file
-    with open(output_file_path, "r") as file:
+    with open(output_file_path, "r", encoding="utf-8") as file:
         saved_output = json.load(file)
 
     # Initialize data structures

main.py

@@ -86,8 +86,8 @@ def save_results_to_file(output_path, agent_config, dataset_config, results, met
     }
 
     # Write to file
-    with open(output_path, "w") as file:
-        json.dump(output_data, file, indent=4)
+    with open(output_path, "w", encoding="utf-8") as file:
+        json.dump(output_data, file, indent=4, ensure_ascii=False)
     logger.info(f"Results saved at {output_path}")
 
 
@@ -180,7 +180,7 @@ def main():
     # Load existing results and initialize tracking variables
     time_cost_list = []
     metrics, results, last_processed_context_id, last_processed_query_id = load_existing_results(
-        output_path, dataset_config, all_query_answer_pairs
+        output_path, dataset_config, all_query_answer_pairs, args.force
     )
 
     # Start evaluation loop - process each context and its associated queries