adds doc searcher v2

app.py

@@ -1,6 +1,7 @@
 import nh3
 from fastapi import FastAPI, Request
 from doc_searcher import DocSearcher
+from doc_searcher_v2 import DocSearcherV2
 from suggestion_searcher import SuggestionSearcher
 from huggingface_hub import login
 from config import HUGGING_FACE_API_KEY, COLLECTION_NAME, API_KEY, COLLECTION_NAME_SUGGESTION
@@ -12,34 +13,21 @@ login(HUGGING_FACE_API_KEY)
 app = FastAPI()
 
 doc_searcher = DocSearcher(collection_name=COLLECTION_NAME)
+doc_searcher_v2 = DocSearcherV2(collection_name=COLLECTION_NAME)
 suggestion_searcher = SuggestionSearcher(collection_name=COLLECTION_NAME_SUGGESTION)
 
 ALLOWED_API_KEY = str(API_KEY)
 
 @app.get("/api/search")
-async def search(q: str):
-    # q: str, type: int, lt: str | None = None, offset: int = 0
+async def search(q: str, type: int, lt: str | None = None, offset: int = 0):
     query = q.lower()
     xss = nh3.clean(query)
-    # data = await doc_searcher.search(text=xss,type=type,law_type=lt,offset=offset)
-    data = await doc_searcher.search_temp(text=xss)
+    data = await doc_searcher.search(text=xss,type=type,law_type=lt,offset=offset)
     return data
 
-@app.get("/api/suggestions")
-async def get_suggestions(q: str, type: int):
+@app.get("/api/v2/search")
+async def v2_search(q: str):
     query = q.lower()
     xss = nh3.clean(query)
-    data = await suggestion_searcher.search(text=xss,type=type)
-    return data
-
-
-# @app.middleware("http")
-# async def api_key_authentication(request: Request, call_next):
-#     api_key = request.headers.get("X-API-KEY")
-#     if api_key != ALLOWED_API_KEY:
-#         return JSONResponse(
-#             status_code=403,
-#             content={"message": "Forbidden."}
-#         )
-
-#     return await call_next(request)
+    data = await doc_searcher_v2.search_temp(text=xss)
+    return data

suggestion_searcher.py → doc_searcher_v2.py

@@ -1,76 +1,48 @@
 from qdrant_client import QdrantClient
-from qdrant_client.models import Filter, FieldCondition, MatchValue
-from fastembed import SparseTextEmbedding, LateInteractionTextEmbedding
+from fastembed import SparseTextEmbedding
 from qdrant_client import QdrantClient, models
+from reranker import Reranker
 from sentence_transformers import SentenceTransformer
-from config import DENSE_MODEL, SPARSE_MODEL, LATE_INTERACTION_MODEL, QDRANT_URL, QDRANT_API_KEY,HUGGING_FACE_API_KEY
+from config import DENSE_MODEL, SPARSE_MODEL, QDRANT_URL, QDRANT_API_KEY
 
-class SuggestionSearcher:
+class DocSearcherV2:
     
     def __init__(self, collection_name):
         self.collection_name = collection_name
-        self.dense_model = SentenceTransformer(DENSE_MODEL,device="cpu",token=HUGGING_FACE_API_KEY)
+        self.reranker = Reranker()
+        self.model = SentenceTransformer("Qwen/Qwen3-Embedding-0.6B",device="cpu")
         self.sparse_model = SparseTextEmbedding(SPARSE_MODEL)
-        self.late_interaction_model = LateInteractionTextEmbedding(LATE_INTERACTION_MODEL)
         self.qdrant_client = QdrantClient(QDRANT_URL,api_key=QDRANT_API_KEY,timeout=30)
     
-    async def search(self, text: str,type:int):
+    async def search_temp(self, text: str):
         
-        dense_query = self.dense_model.encode(text).tolist()
+        queries = [text]
+        dense_query = self.model.encode(text).tolist()
         sparse_query = next(self.sparse_model.query_embed(text))
 
         prefetch = [
                 models.Prefetch(
                     query=dense_query,
-                    params=models.SearchParams(
-                        quantization=models.QuantizationSearchParams(
-                            rescore=False,
-                        ),
-                    ),
-                    using=DENSE_MODEL,
-                    limit=10000
+                    using="Qwen/Qwen3-Embedding-0.6B",
+                    limit=100
                 ),
-                models.Prefetch(
-                    query=models.SparseVector(**sparse_query.as_object()),
-                    params=models.SearchParams(
-                        quantization=models.QuantizationSearchParams(
-                            rescore=False,
-                        ),
-                    ),
-                    using=SPARSE_MODEL,
-                    limit=10000
-                )
         ]
-
-        self.qdrant_client.scroll
+            
         search_result = self.qdrant_client.query_points(
-            collection_name= self.collection_name,
-            query_filter=Filter(
-                must=[
-                    FieldCondition(
-                        key="tip",
-                        match=MatchValue(value=type)
-                    )
-                ]
-            ),
-            search_params=models.SearchParams(
-                hnsw_ef=64, 
-                exact=False, 
-                quantization=models.QuantizationSearchParams(
-                        rescore=True,
-                    ),
-                ),
+            collection_name= "sl-list",
             prefetch=prefetch,
             query=models.FusionQuery(
                 fusion=models.Fusion.RRF,
             ),
             with_payload=True,
-            limit = 5
+            limit = 100,
         ).points
 
         data = []
 
         for hit in search_result:
-            data.append(hit.payload)
+            data.append(hit.payload["tekst"])
+            
+        scores = self.reranker.compute_logits(queries,data)
         
-        return data
+        return scores

reranker.py

@@ -16,7 +16,7 @@ class Reranker:
     def process_inputs(self,pairs):
         prefix = (
             "<|im_start|>system\n"
-            "Proceni da li dati Dokument adekvatno odgovara na Upit na osnovu pravne instrukcije. "
+            "Procijeni da li dati Dokument adekvatno odgovara na Upit na osnovu pravne instrukcije. "
             "Dokument treba da bude relevantan, tačan i u skladu sa važećim pravnim propisima i standardima. "
             "Odgovor mora biti striktno \"da\" ako ispunjava uslove, ili \"ne\" ako ne ispunjava.\n"
             "<|im_end|>\n"