sort?

app.py

@@ -253,7 +253,7 @@ def run_evaluation(dataset, postfix):
     logging.info(f"Number of queries in q_lookups: {len(q_lookups[dataset])}")
     logging.info("Size of all_scores: " + str(len(all_scores)))
     logging.info("Size of psg_indices: " + str(len(psg_indices)))
-    for query, scores, doc_ids in zip(queries, all_scores, psg_indices):
+    for query, scores, doc_ids in zip(queries[current_dataset], all_scores, psg_indices):
         qid = query2qid[dataset][query]
         qid_str = str(qid)
         results[qid_str] = {}