Update app.py

app.py

@@ -96,7 +96,7 @@ async def lifespan(app: FastAPI):
     )
         cross_encoder_model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-base")
         # cross_encoder_model = HuggingFaceCrossEncoder(model_name="BAAI/bge-reranker-large")
-        ml_models["reranker_compressor"] = CrossEncoderReranker(model=cross_encoder_model, top_n=8)
+        ml_models["reranker_compressor"] = CrossEncoderReranker(model=cross_encoder_model, top_n=9)
         ml_models["llm"] = ChatGoogleGenerativeAI(
                                # model="gemini-1.5-pro",
                                model="gemini-2.0-flash",
@@ -225,14 +225,14 @@ async def run_hackrx(req: RunRequest):
     # end_time2 = time.time() - start_time2
     # print(f"vector done: {end_time2}")
     # dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 8})
-    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 12 ,"lambda_mult": 0.5} )
+    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 16 ,"lambda_mult": 0.5} )
     
 
     # Create retrievers using the pre-loaded models from our ml_models dictionary
     keyword_retriever = BM25Retriever.from_documents(chunks)
-    keyword_retriever.k = 8
+    keyword_retriever.k = 12
     # dense_retriever = Chroma.from_documents(documents=chunks, embedding=ml_models["embedder"]).as_retriever()
-    ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.35, 0.65],search_kwargs={"k": 20})
+    ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.35, 0.65],search_kwargs={"k": 16})
     ### to make it faster we are now using our built reranker thats why commenting the code below
     compression_retriever = ContextualCompressionRetriever(
         base_retriever=ensemble_retriever, base_compressor=ml_models["reranker_compressor"]