Upload doc_searcher_v2.py with huggingface_hub

doc_searcher_v2.py

@@ -38,14 +38,46 @@ class DocSearcherV2:
             limit = 100,
         ).points
 
-        data = []
+        # Sačuvaj kompletan hit objekat i tekst za reranking
+        # Kreiraj mapu tekst -> hit za brzo mapiranje
+        text_to_hit = {}
+        texts_for_reranking = []
 
         for hit in search_result:
-            data.append(hit.payload["tekst"])
+            text = hit.payload.get("tekst", "")
+            if text:  # Samo ako postoji tekst
+                text_to_hit[text] = hit
+                texts_for_reranking.append(text)
+        
+        if not texts_for_reranking:
+            return []
             
-        scores = self.reranker.compute_logits(queries,data)
+        # Reranking - vraća top_10 sa (score, query, document) tuple-ovima
+        # queries * len(...) znači da svaki dokument dobija isti query
+        reranked_results = self.reranker.compute_logits(queries * len(texts_for_reranking), texts_for_reranking)
+        
+        # Kombinuj rezultate: mapiraj rerank skorove sa originalnim hit-ovima
+        # reranked_results je lista tuple-ova: [(score, query, document_text), ...]
+        # gde je document_text originalni tekst koji je poslat reranker-u
+        results_with_scores = []
+        for score, query, document_text in reranked_results:
+            # Pronađi originalni hit po tekstu
+            if document_text in text_to_hit:
+                hit = text_to_hit[document_text]
+                # Vrati kompletan payload sa skorom
+                result = {
+                    "score": float(score),
+                    "id": str(hit.id),
+                    "text": document_text,
+                    "payload": hit.payload  # Kompletan payload sa svim podacima
+                }
+                results_with_scores.append(result)
+        
+        # Reranker već vraća sortirano, ali osigurajmo da je sortirano
+        results_with_scores.sort(key=lambda x: x["score"], reverse=True)
         
-        return scores
+        # Vrati top rezultate (reranker već vraća top 10)
+        return results_with_scores
     
     async def search_keyword(self, text: str):
         sparse_query = next(self.sparse_model.query_embed(text))