Update app.py

app.py

@@ -1,8 +1,10 @@
+import os
 import gradio as gr
-from langchain.chains import RagChain
 from langchain.vectorstores import Chroma
 from transformers import RagTokenizer, RagSequenceForGeneration
 from sentence_transformers import SentenceTransformer
+from langchain.chains.question_answering import load_qa_chain
+from langchain.llms import HuggingFaceLLM
 
 #Konstanten
 ANTI_BOT_PW = os.getenv("CORRECT_VALIDATE")
@@ -20,11 +22,16 @@ model = RagSequenceForGeneration.from_pretrained("facebook/rag-sequence-nq", use
 # Verbindung zur Chroma DB und Laden der Dokumente
 chroma_db = Chroma(embedding_model=embedding_model, persist_directory = PATH_WORK + CHROMA_DIR)
 
+# Erstellung eines HuggingFaceLLM Modells
+llm = HuggingFaceLLM(model=model, tokenizer=tokenizer)
+
+
+
 # Erstellen eines eigenen Retrievers mit Chroma DB und Embeddings
-retriever = chroma_db.as_retriever()
+#retriever = chroma_db.as_retriever()
 
 # Erstellung der RAG-Kette mit dem benutzerdefinierten Retriever
-rag_chain = RagChain(model=model, retriever=retriever, tokenizer=tokenizer, vectorstore=chroma_db)
+#rag_chain = RagChain(model=model, retriever=retriever, tokenizer=tokenizer, vectorstore=chroma_db)
 #############################################
 
 
@@ -45,23 +52,25 @@ def document_retrieval_chroma2():
 	
 	
 	
-def get_rag_response(prompt):
+def get_rag_response(question):
 	global rag_chain
-	#rag-chain nutzen, um Antwort zu generieren
-	result = rag_chain({"Frage: " : prompt})
-	
-	#relevante Dokumente extrahieren
-	docs = result['docs']
-	passages = [doc['text'] for doc in docs]
-	links = [doc['url'] for doc in docs]
-	
-	#Antwort generieren
-	answer = result['output']
-	response = {
-			"answer" : answer,
-			"documents" : [{"link" : link, "passage" : passage} for link, passage in zip(links, passages)]
-			}
-	return response
+
+    # Abfrage der relevanten Dokumente aus Chroma DB
+    docs = chroma_db.search(question, top_k=5)
+    passages = [doc['text'] for doc in docs]
+    links = [doc.get('url', 'No URL available') for doc in docs]
+
+    # Generieren der Antwort
+    answer = llm(question, docs)
+    
+    # Zusammenstellen der Ausgabe
+    response = {
+        "answer": answer,
+        "documents": [{"link": link, "passage": passage} for link, passage in zip(links, passages)]
+    }
+    
+    return response
+
 	
 	
 def chatbot_response (user_input, chat_history=[]):