Update app.py

app.py

@@ -13,8 +13,8 @@ from huggingface_hub import snapshot_download, hf_hub_download
 
 # from prompts import CONDENSE_QUESTION_PROMPT
 
-repo_name = "IlyaGusev/saiga2_13b_gguf"
-model_name = "model-q4_K.gguf"
+repo_name = "IlyaGusev/saiga2_7b_gguf"
+model_name = "model-q2_K.gguf"
 
 #snapshot_download(repo_id=repo_name, local_dir=".", allow_patterns=model_name)
 
@@ -50,17 +50,17 @@ def get_vectorstore(text_chunks):
 
 
 def get_conversation_chain(vectorstore, model_name):
-    #llm = LlamaCpp(model_path=model_name,
-    #               temperature=0.1,
-    #               top_k=30,
-    #               top_p=0.9,
-    #               streaming=True,
-    #               n_ctx=2048,
-    #               n_parts=1,
-    #               echo=True
-    #               )
-
-    llm = ChatOpenAI()
+    llm = LlamaCpp(model_path=model_name,
+                   temperature=0.1,
+                   top_k=30,
+                   top_p=0.9,
+                   streaming=True,
+                   n_ctx=2048,
+                   n_parts=1,
+                   echo=True
+                   )
+
+    #llm = ChatOpenAI()
     
     memory = ConversationBufferMemory(memory_key='chat_history', input_key='question', output_key='answer', return_messages=True)
     
@@ -68,7 +68,7 @@ def get_conversation_chain(vectorstore, model_name):
                                                                # condense_question_prompt=CONDENSE_QUESTION_PROMPT,
                                                                retriever=vectorstore.as_retriever(),
                                                                memory=memory,
-                                                               #return_source_documents=True
+                                                               return_source_documents=True
                                                                )
 
     return conversation_chain
@@ -79,9 +79,12 @@ def handle_userinput(user_question):
 
     st.session_state.chat_history = response['chat_history']
 
-    #st.session_state.retrieved_text = response['source_documents'][0]
+    st.session_state.retrieved_text = response['source_documents']
 
-    #print(response['source_documents'][0])
+    for n in enumerate(st.session_state.retrieved_text):
+        st.write(response['source_documents'][n])
+
+    print(response['source_documents'][0])
 
     for i, message in enumerate(st.session_state.chat_history):
         if i % 2 == 0: