Update app.py

app.py

@@ -61,7 +61,9 @@ 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",
+        ml_models["embedder"] = HuggingFaceEmbeddings(
+                                                      # model_name="BAAI/bge-base-en-v1.5", #better but lil slower
+                                                      model_name="intfloat/e5-large-v2", #lil faster 
                                                       encode_kwargs={
                                                           "batch_size": 64,
                                                           # "normalize_embeddings": True
@@ -164,12 +166,12 @@ 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": 7 ,"lambda_mult": 0.5})
+    dense_retriever = vectorstore.as_retriever(search_type="mmr",search_kwargs={"k": 5 ,"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 = 4
     # 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