Update app.py

app.py

@@ -1,6 +1,7 @@
 import gradio as gr
 from operator import itemgetter
 import os
+import pandas as pd 
 
 from langchain_community.vectorstores import FAISS
 from langchain_core.output_parsers import StrOutputParser
@@ -52,12 +53,15 @@ retriever = vectorstore.as_retriever()
 
 qa = RetrievalQA.from_chain_type(
 llm=hf, chain_type="stuff", retriever=retriever, return_source_documents=False)
+queries=pd.read_csv('./interactions/queries.csv')
 
 def greet(Question):
     answer = qa({"query": Question})
 
     pa=[a.split("Helpful Answer: ") for a in answer.get('result').split('\n') if "Helpful Answer" in a]
-
+    new=pd.DataFrame({'query':Question,'response':pa[0][-1]})
+    queries.append(new)
+    queries.to_csv('./interactions/queries.csv')
     return pa[0][-1]
 
 if __name__ == "__main__":
@@ -67,8 +71,11 @@ if __name__ == "__main__":
     description = """
     <img src="https://superagi.com/wp-content/uploads/2023/10/Introduction-to-RAGA-Retrieval-Augmented-Generation-and-Actions-1200x600.png.webp" width=100%>
     <br>
-    Demo using TinyLlama, a chat model finetuned on top of TinyLlama/TinyLlama-1.1B-intermediate-step-1431k-3T. This space demonstrate application of RAG on a small model and its effectiveness, I used it because of the space constraint. The current space runs on mere <b>2GB of RAM</b>, hence there is some delay in generating output. Test this to your hearts content and let me know your thoughts, I will keep updating this space with tiny improvements on architecture and design
+    Demo using Vector store-backed retriever. This space demonstrate application of RAG on a small model and its effectiveness, I used small model because of the space constraint. The current space runs on mere <b>2GB of RAM</b>, hence there is some delay in generating output. Test this to your hearts content and let me know your thoughts, I will keep updating this space with tiny improvements on architecture and design
     <ul>
+        <li>model: TinyLlama/TinyLlama-1.1B-Chat-v1.0</li>
+        <li></li>
+        
         <li>update1: This space now does not create a faiss index on build, it uses a locally saved faiss index</li>
         <li>update2: This space now uses google/gemma-1.1-2b-it model to generate output, reduces the response time to 1/3rd</li>
     </ul>
@@ -77,6 +84,8 @@ if __name__ == "__main__":
     <ul>You can ask questions like - 
         <li>What is langchain framework?</li>
         <li>What is Action Agent?</li>
+        <li>What are forms of memory implementation in langchain</li>
+        <li>What is question answering from documents</li>
     </ul>
     Go through this paper here to find more about langchain and then test how this solution performs. <a href='https://www.researchgate.net/publication/372669736_Creating_Large_Language_Model_Applications_Utilizing_LangChain_A_Primer_on_Developing_LLM_Apps_Fast' target='_blank'>This paper is the data source for this solution</a>
     Have you already used RAG? feel free to suggest improvements