Update app.py

app.py

@@ -229,15 +229,15 @@ 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": 10 ,"lambda_mult": 0.8} ) #prev 12
+    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 12 ,"lambda_mult": 0.8} ) #prev 12
     # dense_retriever = vectorstore.as_retriever(search_type="similarity" ,search_kwargs={"k": 11} )
     
 
     # Create retrievers using the pre-loaded models from our ml_models dictionary
-    # keyword_retriever = BM25Retriever.from_documents(chunks)
-    # keyword_retriever.k = 7
+    keyword_retriever = BM25Retriever.from_documents(chunks)
+    keyword_retriever.k = 7
     # dense_retriever = Chroma.from_documents(documents=chunks, embedding=ml_models["embedder"]).as_retriever()
-    # ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.6, 0.4],search_kwargs={"k": 9})
+    ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.6, 0.4],search_kwargs={"k": 9})
     ### 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"]
@@ -245,7 +245,7 @@ async def run_hackrx(req: RunRequest):
 
     # Define the RAG chain using pre-loaded components
     hybrid_rag_chain = (
-        {"context": itemgetter("full_query") | dense_retriever, "full_query": itemgetter("full_query")}
+        {"context": itemgetter("full_query") | ensemble_retriever, "full_query": itemgetter("full_query")}
         | ml_models["prompt_template"]
         | ml_models["llm"]
     )