Update app.py

app.py

@@ -22,7 +22,7 @@ from langchain_huggingface import HuggingFaceEmbeddings # Correct new import
 # from langchain_chroma import Chroma
 from langchain_community.retrievers import BM25Retriever
 from langchain.retrievers import EnsembleRetriever, ContextualCompressionRetriever
-from langchain.retrievers.document_compressors import CrossEncoderReranker
+from langchain.retrievers.document_compressors import CrossEncoderReranker , DocumentCompressorPipeline
 from langchain_community.cross_encoders import HuggingFaceCrossEncoder
 from langchain.prompts import PromptTemplate
 
@@ -157,18 +157,21 @@ async def run_hackrx(req: RunRequest):
         embedding=ml_models["embedder"]
     )
 
-    dense_retriever = vectorstore.as_retriever(search_kwargs={"k": 4})
+    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 8})
     
 
     # Create retrievers using the pre-loaded models from our ml_models dictionary
     keyword_retriever = BM25Retriever.from_documents(chunks)
-    keyword_retriever.k = 3
+    keyword_retriever.k = 5
     # dense_retriever = Chroma.from_documents(documents=chunks, embedding=ml_models["embedder"]).as_retriever()
     ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.4, 0.6])
     
-    compression_retriever = ContextualCompressionRetriever(
+    compression_retriever = DocumentCompressorPipeline(
         base_retriever=ensemble_retriever, base_compressor=ml_models["reranker_compressor"]
     )
+    # compression_retriever = ContextualCompressionRetriever(
+    #     base_retriever=ensemble_retriever, base_compressor=ml_models["reranker_compressor"]
+    # )
 
     # Define the RAG chain using pre-loaded components
     hybrid_rag_chain = (