diff --git a/local_evaluation.py b/local_evaluation.py
index 1723e2a..db751ec 100644
--- a/local_evaluation.py
+++ b/local_evaluation.py
@@ -182,7 +182,7 @@ def generate_predictions(dataset_path, participant_model):
     return queries, ground_truths, predictions
 
 
-def evaluate_predictions(queries, ground_truths_list, predictions, evaluation_model_name):
+def evaluate_predictions(queries, ground_truths_list, predictions, evaluation_model_name, openai_client):
     """
     Evaluates the predictions generated by a model against ground truth answers.
     
@@ -199,7 +199,7 @@ def evaluate_predictions(queries, ground_truths_list, predictions, evaluation_mo
 
     if "chat" in evaluation_model_name.lower():
         # now we are using chatgpt
-        openai_client = OpenAI()
+        # openai_client = OpenAI()
         n_miss, n_correct = 0, 0
         system_message = get_system_message()
 
@@ -211,6 +211,7 @@ def evaluate_predictions(queries, ground_truths_list, predictions, evaluation_mo
             # trim prediction to 75 tokens using Llama2 tokenizer
             prediction = trim_predictions_to_max_token_length(prediction)
             prediction = prediction.strip()
+            prediction_lowercase = prediction.lower()
 
             if "i don't know" in prediction_lowercase:
                 n_miss += 1
@@ -220,7 +221,6 @@ def evaluate_predictions(queries, ground_truths_list, predictions, evaluation_mo
 
             for ground_truth in ground_truths:
                 ground_truth_lowercase = ground_truth.lower()
-                prediction_lowercase = prediction.lower()
                 messages = [
                     {"role": "system", "content": system_message},
                     {