update

app.py

@@ -41,14 +41,14 @@ def search(query: str, reranking: bool = False, top_k: int = 100):
     hits = util.semantic_search(question_embedding, corpus_embeddings, top_k=top_k)
     hits = hits[0]  # Get the hits for the first query
 
-    top_k_child_passages = [child_passages[hit["corpus_id"]] for hit in hits]
-    top_k_child_passage_ids = [hit["corpus_id"] for hit in hits]
+    top_k_child_passages = [child_passages[hit["corpus_id"]] for hit in hits][:20]
+    top_k_child_passage_ids = [hit["corpus_id"] for hit in hits][:20]
 
     ##### Re-Ranking #####
     # Now, score all retrieved passages with the cross_encoder
     if reranking:
         colbert_scores: list[dict[str, Any]] = colbert.rerank(
-            query=query, documents=top_k_child_passages, top_k=100
+            query=query, documents=top_k_child_passages, top_k=20
         )
 
         # Reorder child passage ids based on the reranking