Update app.py

app.py

@@ -7,44 +7,81 @@ from langchain.chains import RetrievalQA
 from langchain_community.llms import HuggingFaceHub
 
 def make_vectorstore(embeddings):
-    
-    loader = PyPDFDirectoryLoader("data")   
+    loader = PyPDFDirectoryLoader("/content/data") 
     documents = loader.load()
-    text_splitter = CharacterTextSplitter(chunk_size=1400, chunk_overlap=0)
+    text_splitter = CharacterTextSplitter(chunk_size=400, chunk_overlap=0)
     texts = text_splitter.split_documents(documents)
     docsearch = FAISS.from_documents(texts, embeddings)
     
     return docsearch
 
-def get_qa(vectorstore, llmb):
+def get_conversation(vectorstore, model):
 
-    qa = RetrievalQA.from_llm(
-        llm=llmb,
-        chain_type="stuff", 
+    conversation_chain = RetrievalQA.from_llm(
+        llm=model,
+        # chain_type="stuff", 
         retriever=vectorstore.as_retriever())
     
-    return qa
+    return conversation_chain
 
-def get_response(qa, query):
-    
-    response = qa.run(query)
-    
+def get_response(conversation_chain, query):
+    # get the response
+    response = conversation_chain.invoke(query)
     return response
 
+def response_formatter(resp_list):
+    queries = []
+    responses = []
+    for resp in resp_list:
+        # find the '\nQuestion: ' and '\nHelpful Answer: ' and take the text right in front of them a new list of query and responses
+        content = resp["result"]
+        # find '\nQuestion: ' in the text
+        question =  content.split('\nQuestion: ')[1].split('\nHelpful Answer: ')[0]
+        queries.append(question)
+        # find '\nHelpful Answer: ' in the text
+        answer = content.split('\nHelpful Answer: ')[1]
+        responses.append(answer)
+    return queries, responses 
+
 def main():
-    
     st.title("BetterZila RAG Enabled LLM")
-    llm = HuggingFaceHub(repo_id="google/flan-t5-xxl", model_kwargs={"temperature":0.5, "max_length":512}, huggingfacehub_api_token = st.secrets["hf_token"])
-    embeddings = HuggingFaceInstructEmbeddings(model_name="google/t5-v1_1-xl", model_kwargs = {'device': 'cpu'})
-    vectorstore = make_vectorstore(embeddings)
-    qa = get_qa(vectorstore, llm)
-    queries = ["Can you give me an example from history where the enemy was crushed totally from the book?", "What's the point of making myself less accessible?", "Can you tell me the story of Queen Elizabeth I from this 48 laws of power book?"]
+    
+    # a sidebar taht titles a disclaimer and a description of the app
+    st.sidebar.title("About")
+    st.sidebar.info("This app is a demo of BetterZila RAG Enabled LLM")
+    # write a paragraph in the sidebar
+    st.sidebar.write("This app is a demo of BetterZila RAG Enabled LLM")
+    
+    response_list=[]
+    
+    print("Downloading Embeddings Model")
+    with st.spinner('Downloading Embeddings Model...'):
+        embeddings = HuggingFaceInstructEmbeddings(model_name="hkunlp/instructor-base", model_kwargs = {'device': 'cpu'})
+    
+    print("Loading LLM from HuggingFace")
+    with st.spinner('Loading LLM from HuggingFace...'):
+        llm = HuggingFaceHub(repo_id="HuggingFaceH4/zephyr-7b-beta", model_kwargs={"temperature":0.7, "max_new_tokens":512, "top_p":0.95, "top_k":50})
+    
+    print("Creating Vector Database of PDF file content")
+    with st.spinner('Creating Vector Database of PDF file content...'):
+        vectorstore = make_vectorstore(embeddings)
+    
+    print("Intializing LLM for inference with source material")
+    with st.spinner('Intializing LLM for inference with source material...'):
+        conversation_chain = get_conversation(vectorstore, llm)
+    
+    queries = ["Can you give me an example from history where the enemy was crushed totally from the book?",
+               "What's the point of making myself less accessible?",
+               "Can you tell me the story of Queen Elizabeth I from this 48 laws of power book?"]
+    
     for query in queries:
-        st.subheader(f"Query: {query}")
-        response = get_response(qa, query)
-        st.write(query)
-        st.write(response)
-    st.success("Responses generated!")
+        response = get_response(conversation_chain, query)
+        response_list.append(response)
+    queries, responses = response_formatter(response_list)
+    for i in range(len(queries)):
+        st.write("Query: ", queries[i])
+        st.write("Response: ", responses[i])
+        st.write("--------------------------------------------------")
     
 if __name__ == "__main__":
     main()