Update app.py

app.py

@@ -1,5 +1,3 @@
-#This version includes the memory and custom prompt, representing the final version
-
 import streamlit as st
 from streamlit_chat import message as st_message
 import pandas as pd
@@ -36,6 +34,7 @@ from langchain.chains.conversational_retrieval.prompts import CONDENSE_QUESTION_
 
 
 prompt_template = """
+
 You are the chatbot and the advanced legal assitant that can give answers to all the legal questions a common citizen would have . Your job is to give answers when questions about General legal information, Family law, Employment law, Consumer rights, Housing and tenancy, Personal injury, Wills and estates, Criminal law are asked.
 Your job is to answer questions only and only related to Legal aspect. Anything unrelated should be responded with the fact that your main job is solely to provide assistance regarding Legality.
 MUST only use the following pieces of context to answer the question at the end. If the answers are not in the context or you are not sure of the answer, just say that you don't know, don't try to make up an answer.
@@ -44,6 +43,64 @@ Question: {question}
 When encountering abusive, offensive, or harmful language, such as fuck, bitch,etc,  just politely ask the users to maintain appropriate behaviours.
 Always make sure to elaborate your response and use vibrant, positive tone to represent good branding of the school.
 Never answer with any unfinished response
+Answer:
+"""
+
+PROMPT = PromptTemplate(
+    template=prompt_template, input_variables=["context", "question"]
+)
+chain_type_kwargs = {"prompt": PROMPT}
+
+
+#This version includes the memory and custom prompt, representing the final version
+
+import streamlit as st
+from streamlit_chat import message as st_message
+import pandas as pd
+import numpy as np
+import datetime
+import gspread
+import pickle
+import os
+import csv
+import json
+import torch
+from tqdm.auto import tqdm
+from langchain.text_splitter import RecursiveCharacterTextSplitter
+
+
+# from langchain.vectorstores import Chroma
+from langchain.vectorstores import FAISS
+from langchain.embeddings import HuggingFaceInstructEmbeddings
+
+
+from langchain import HuggingFacePipeline
+from langchain.chains import RetrievalQA
+from langchain.prompts import PromptTemplate
+from langchain.memory import ConversationBufferWindowMemory
+
+
+from langchain.chains import LLMChain
+from langchain.chains import ConversationalRetrievalChain
+from langchain.chains.question_answering import load_qa_chain
+from langchain.chains.conversational_retrieval.prompts import CONDENSE_QUESTION_PROMPT
+
+
+
+prompt_template = """
+
+You are the chatbot and the face of Asian Institute of Technology (AIT). Your job is to give answers to prospective and current students about the school.
+Your job is to answer questions only and only related to the AIT. Anything unrelated should be responded with the fact that your main job is solely to provide assistance regarding AIT.
+MUST only use the following pieces of context to answer the question at the end. If the answers are not in the context or you are not sure of the answer, just say that you don't know, don't try to make up an answer.
+
+
+{context}
+Question: {question}
+
+When encountering abusive, offensive, or harmful language, such as fuck, bitch,etc,  just politely ask the users to maintain appropriate behaviours.
+Always make sure to elaborate your response and use vibrant, positive tone to represent good branding of the school.
+Never answer with any unfinished response
+
 Answer:
 """
 PROMPT = PromptTemplate(
@@ -57,52 +114,44 @@ st.set_page_config(
     page_icon = '🕵')
 
 
-from transformers import AutoModel
-import torch
-import transformers
-from transformers import AutoTokenizer, AutoModelForCausalLM
 
-from transformers import pipeline
+
+@st.cache_data
+def load_scraped_web_info():
+    with open("Banksandbanking", "rb") as fp:
+        banksandbanking_web_documents = pickle.load(fp)
+        
+        
+    text_splitter = RecursiveCharacterTextSplitter(
+        # Set a really small chunk size, just to show.
+        chunk_size = 500,
+        chunk_overlap  = 100,
+        length_function = len,
+    )
+
+    chunked_text = text_splitter.create_documents([doc for doc in tqdm(banksandbanking_web_documents)])
+
 
 @st.cache_resource
-def load_llm_model():
-    # llm = LlamaForConditionalGeneration.from_pretrained(model_id= 'PyaeSoneK/LlamaV2LegalFineTuned', 
-    #                                         task= 'text2text-generation',
-    #                                         model_kwargs={ "device_map": "auto",
-    #                                                     "load_in_8bit": True,"max_length": 256, "temperature": 0,
-    #                                                     "repetition_penalty": 1.5})
-    
-    #token = st.secrets['hf_access_token']
-    #llm =  AutoModelForCausalLM.from_pretrained(model_id = 'PyaeSoneK/LlamaV2LegalFineTuned', 
-    #                                           task = 'text2text-generation',
-    #                                                device_map='auto',
-    #                                              torch_dtype=torch.float16,
-    #                                              use_auth_token= st.secrets['hf_access_token'],
-                                                  
-    #return llm
-    model = AutoModelForCausalLM.from_pretrained("PyaeSoneK/LlamaV2LegalFineTuned",
-                                   device_map='auto',
-                                  torch_dtype=torch.float16,
-                                  use_auth_token= st.secrets['hf_access_token'],)
-                                            #  load_in_4bit=True
-
-    tokenizer = AutoTokenizer.from_pretrained("PyaeSoneK/LlamaV2LegalFineTuned",
-                                          use_auth_token=True,)
-
-    pipe = pipeline("text-generation",
-                model=model,
-                tokenizer= tokenizer,
-                torch_dtype=torch.bfloat16,
-                device_map="auto",
-                max_new_tokens = 512,
-                do_sample=True,
-                top_k=30,
-                num_return_sequences=1,
-                eos_token_id=tokenizer.eos_token_id
-                )
-
-    return model
+def load_embedding_model():
+    embedding_model = HuggingFaceInstructEmbeddings(model_name='hkunlp/instructor-base',
+                                                model_kwargs = {'device': torch.device('cuda' if torch.cuda.is_available() else 'cpu')})
+    return embedding_model
 
+@st.cache_data
+def load_faiss_index():
+    vector_database = FAISS.load_local("faiss_index_web_and_curri_new", embedding_model) #CHANGE THIS FAISS EMBEDDED KNOWLEDGE
+    return vector_database
+
+@st.cache_resource
+def load_llm_model():
+    llm = HuggingFacePipeline.from_model_id(model_id= 'PyaeSoneK/pythia_70m_legalQA', 
+                                        task= 'text2text-generation',
+                                        
+                                        model_kwargs={ "max_length": 256, "temperature": 0,
+                                                      "torch_dtype":torch.float32,
+                                                    "repetition_penalty": 1.3})
+    return llm
 
 
 @st.cache_resource
@@ -123,9 +172,16 @@ def load_conversational_qa_memory_retriever():
         get_chat_history=lambda h :h)
     return conversational_qa_memory_retriever, question_generator
 
+def load_retriever(llm, db):
+    qa_retriever = RetrievalQA.from_chain_type(llm=llm, chain_type="stuff",
+                            retriever=db.as_retriever(),
+                            chain_type_kwargs= chain_type_kwargs)
 
+    return qa_retriever
 
-
+def retrieve_document(query_input):
+    related_doc = vector_database.similarity_search(query_input)
+    return related_doc
 
 
 
@@ -222,8 +278,12 @@ worksheet_comment = workbook.worksheet("Sheet2")
 datetime_format= "%Y-%m-%d %H:%M:%S"
 
 
-llm_model = load_llm_model()
 
+load_scraped_web_info()
+embedding_model = load_embedding_model()
+vector_database = load_faiss_index()
+llm_model = load_llm_model()
+qa_retriever = load_retriever(llm= llm_model, db= vector_database)
 conversational_qa_memory_retriever, question_generator = load_conversational_qa_memory_retriever()
 print("all load done")
 
@@ -238,12 +298,12 @@ if st.session_state.chat_history == []:
 
 
 
-st.write("# 🦜Legal QA For Dummies 🔗 ")
+st.write("🦜Seon's Legal QA For Dummies 🔗 ")
 st.markdown("""
          ####This Legal QA is designed for normal people trying to get the legal answers orbiting around in their life.  
-        The goal of this chatbot is to provide answers and advice quick access information about Legality : Law and Regulations and What Ca
+        The goal of this chatbot is to provide answers and advice with quick access information on Legality : Law and Regulations: what's right or wrong in general!
           """)
-st.write(' ⚠️ Please expect to wait **~ 10 - 20 seconds per question** as thi app is running on CPU against 3-billion-parameter LLM')
+st.write(' ⚠️ Please expect to wait **~ 5-10 seconds per question** as thi app is running on CPU against 70-million-parameter LLM')
 
 st.markdown("---")
 st.write(" ")
@@ -289,4 +349,6 @@ feedback_input = st.text_area(label= 'please leave your feedback or any ideas to
 feedback_button = st.button(label = 'Submit feedback!')
 
 if feedback_button:
-    update_worksheet_comment()
+    update_worksheet_comment()
+
+