Update app.py

app.py

@@ -28,8 +28,8 @@ from langchain.retrievers import ContextualCompressionRetriever
 from langchain.retrievers.document_compressors import CrossEncoderReranker 
 from langchain_community.cross_encoders import HuggingFaceCrossEncoder
 from langchain.prompts import ChatPromptTemplate
-from langchain_nvidia_ai_endpoints.embeddings import NVIDIAEmbeddings
-from langchain_nvidia_ai_endpoints.reranking import  NVIDIARerank
+# from langchain_nvidia_ai_endpoints.embeddings import NVIDIAEmbeddings
+# from langchain_nvidia_ai_endpoints.reranking import  NVIDIARerank
 
 import os
 from sentence_transformers import SentenceTransformer
@@ -68,19 +68,19 @@ async def lifespan(app: FastAPI):
             raise RuntimeError("CRITICAL: Missing GOOGLE_API_KEY in environment secrets!")
     
         # Load models into the shared dictionary
-        # ml_models["embedder"] = HuggingFaceEmbeddings(
-        #                                               model_name="BAAI/bge-base-en-v1.5", #better but lil slower
-        #                                               # model_name="intfloat/e5-large-v2", #lil faster  but  dont know response is slow 
-        #                                               encode_kwargs={
-        #                                                   "batch_size": 64,
-        #                                                   # "normalize_embeddings": True
-        #                                               }
-        ml_models["embedder"] = NVIDIAEmbeddings(model="nvidia/nv-embedqa-e5-v5", nvidia_api_key=nvidia_api_key)
+        ml_models["embedder"] = HuggingFaceEmbeddings(
+                                                      model_name="BAAI/bge-base-en-v1.5", #better but lil slower
+                                                      # model_name="intfloat/e5-large-v2", #lil faster  but  dont know response is slow 
+                                                      encode_kwargs={
+                                                          "batch_size": 64,
+                                                          # "normalize_embeddings": True
+                                                      }
+        # ml_models["embedder"] = NVIDIAEmbeddings(model="nvidia/nv-embedqa-e5-v5", nvidia_api_key=nvidia_api_key)
                                                     
     )
         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=6)
+        ml_models["reranker_compressor"] = CrossEncoderReranker(model=cross_encoder_model, top_n=8)
         ml_models["llm"] = ChatGoogleGenerativeAI(
                                # model="gemini-1.5-pro",
                                model="gemini-2.0-flash",
@@ -174,12 +174,12 @@ async def run_hackrx(req: RunRequest):
     )
 
     # dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 8})
-    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 8 ,"lambda_mult": 0.5})
+    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 12 ,"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 = 5
+    keyword_retriever.k = 8
     # 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])
     ### to make it faster we are now using our built reranker thats why commenting the code below