Update app.py

app.py

@@ -99,7 +99,8 @@ async def lifespan(app: FastAPI):
         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",
+                               # model="gemini-2.0-flash",
+                               model="gemini-2.0-pro",
                                api_key=GOOGLE_API_KEY,
                                temperature=0.1,
                                max_output_tokens=300
@@ -201,8 +202,8 @@ async def run_hackrx(req: RunRequest):
             return JSONResponse({"error": "No documents could be processed."}, status_code=400)    
     end_time = time.time() - start_time
     print(f"chunking done: {end_time}")
-    if not chunks:
-        return JSONResponse({"error": "No documents could be processed."}, status_code=400)
+    # if not chunks:
+    #     return JSONResponse({"error": "No documents could be processed."}, status_code=400)
 
 
     start_time2 = time.time()
@@ -227,22 +228,22 @@ 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": 16 ,"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 = 12
+    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],search_kwargs={"k": 16})
+    ensemble_retriever = EnsembleRetriever(retrievers=[keyword_retriever, dense_retriever], weights=[0.35, 0.65],search_kwargs={"k": 10})
     ### 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"]
-    )
+    # compression_retriever = ContextualCompressionRetriever(
+    #     base_retriever=ensemble_retriever, base_compressor=ml_models["reranker_compressor"]
+    # )
 
     # Define the RAG chain using pre-loaded components
     hybrid_rag_chain = (
-        {"context": itemgetter("full_query") | compression_retriever, "full_query": itemgetter("full_query")}
+        {"context": itemgetter("full_query") | ensemble_retriever, "full_query": itemgetter("full_query")}
         | ml_models["prompt_template"]
         | ml_models["llm"]
     )