from gradio_client import Client from langchain.document_loaders.text import TextLoader from langchain.text_splitter import RecursiveCharacterTextSplitter from langchain.schema import Document from langchain.embeddings import HuggingFaceEmbeddings from langchain import PromptTemplate from langchain.callbacks.streaming_stdout import StreamingStdOutCallbackHandler from langchain.callbacks.manager import CallbackManager from langchain.vectorstores import FAISS from langchain.chains import RetrievalQA from langchain.memory import ConversationBufferMemory from langchain.chains import ConversationalRetrievalChain from huggingface_hub import hf_hub_download from langchain.llms import LlamaCpp from langchain.chains import LLMChain import time import streamlit as st class MyBot: def __init__(self, text_file, model_id, model_basename): self.text_file = text_file self.model_id = model_id self.model_basename = model_basename self.loader = TextLoader(self.text_file) self.pages = self.loader.load() self.chunks_text = self.split_text(self.pages) self.docs_text = [doc.page_content for doc in self.chunks_text] self.embedding = HuggingFaceEmbeddings(model_name='sentence-transformers/all-MiniLM-L6-v2') self.VectorStore = FAISS.from_texts(self.docs_text, embedding=self.embedding) self.model_path = self.download_model(self.model_id, self.model_basename) self.callback_manager = CallbackManager([StreamingStdOutCallbackHandler()]) self.llm = self.init_llm(self.model_path, self.callback_manager) self.memory = ConversationBufferMemory( memory_key="chat_history", return_messages=True, input_key='question', output_key='answer' ) self.qa = self.init_qa(self.llm, self.VectorStore, self.memory) def split_text(self, documents): text_splitter = RecursiveCharacterTextSplitter( chunk_size=1000, chunk_overlap=150, length_function=len, add_start_index=True, ) chunks = text_splitter.split_documents(documents) return chunks def download_model(self, model_id, model_basename): model_path = hf_hub_download( repo_id=model_id, filename=model_basename, resume_download=True, ) print("model_path : ", model_path) return model_path def init_llm(self, model_path, callback_manager): CONTEXT_WINDOW_SIZE = 1500 MAX_NEW_TOKENS = 2000 N_BATCH = 512 n_gpu_layers = 40 kwargs = { "model_path": model_path, "n_ctx": CONTEXT_WINDOW_SIZE, "max_tokens": MAX_NEW_TOKENS, "n_batch": N_BATCH, "n_gpu_layers": n_gpu_layers, "callback_manager": callback_manager, "verbose":True, } llm = LlamaCpp(**kwargs) return llm def init_qa(self, llm, VectorStore, memory): qa = ConversationalRetrievalChain.from_llm( llm, chain_type="stuff", retriever=VectorStore.as_retriever(search_kwargs={"k": 5}), memory=memory, return_source_documents=True, verbose=False, ) return qa def translate(self, text, source,target): client = Client("https://facebook-seamless-m4t-v2-large.hf.space/--replicas/2bmbx/") result = client.predict( text, source, target, api_name="/t2tt" ) #--------------------------------------------------------- # Set page config st.set_page_config( page_title="🤖💼 🇲🇦 Financial advisor is Here", page_icon="🤖", layout="wide", initial_sidebar_state="expanded", ) # Set Streamlit theme # Define the custom CSS custom_css = """ """ # Add the custom CSS to the app st.markdown(custom_css, unsafe_allow_html=True) # Replicate Credentials with st.sidebar: st.title('Mokawil.AI is Here 🤖💼') st.markdown('🤖 An AI-powered advisor designed to assist founders (or anyone aspiring to start their own company) with various aspects of business in Morocco. This includes legal considerations, budget planning, strategies for success, and much more.') selected_language = st.sidebar.selectbox("Select Language", ["English", "Darija"], index=0) # English is the default # Store LLM generated responses if "messages" not in st.session_state.keys(): st.session_state.messages = [{"role": "assistant", "content": "How may I assist you today?"}] # Display or clear chat messages for message in st.session_state.messages: if message["role"] == "user" : with st.chat_message(message["role"], avatar="user.png"): st.write(message["content"]) else : with st.chat_message(message["role"], avatar="logo.png"): st.write(message["content"]) # Create an instance of LangChain lc = MyBot("Data_blog.txt", "TheBloke/Mistral-7B-OpenOrca-GGUF", "mistral-7b-openorca.Q4_K_M.gguf") # Use the instance methods in your Streamlit application def clear_chat_history(): lc.memory.clear() lc.qa = lc.init_qa(lc.llm, lc.VectorStore, lc.memory) st.session_state.messages = [{"role": "assistant", "content": "How may I assist you today?"}] def generate_llm_response(prompt_input): res = lc.qa(f'''{prompt_input}''') if selected_language == "Darija": translated_response = lc.translate(res['answer']) return translated_response else: return res['answer'] # User-provided prompt if prompt := st.chat_input("Cities to start my buisiness in finance?"): if selected_language == "Darija": tprompt = translate(str(prompt),"Moroccan Arabic","English") else: tprompt = prompt st.session_state.messages.append({"role": "user", "content": tprompt}) with st.chat_message("user", avatar="user.png"): st.write(prompt) # Generate a new response if last message is not from assistant if st.session_state.messages[-1]["role"] != "assistant": with st.chat_message("assistant", avatar="logo.png"): with st.spinner("Thinking..."): response = generate_llm_response(st.session_state.messages[-1]["content"]) placeholder = st.empty() full_response = '' for item in response: full_response += item placeholder.markdown(full_response) time.sleep(0.05) placeholder.markdown(full_response) message = {"role": "assistant", "content": full_response} st.session_state.messages.append(message) # Example prompt with st.sidebar : st.title('Examples :') def promptExample1(): prompt = "How can I start my company in Morocco?" st.session_state.messages.append({"role": "user", "content": prompt}) # Example prompt def promptExample2(): prompt = "What are some recommended cities for starting a business in the finance sector?" st.session_state.messages.append({"role": "user", "content": prompt}) # Example prompt def promptExample3(): prompt = "What is the estimated amount of money I need to start my company?" st.session_state.messages.append({"role": "user", "content": prompt}) st.sidebar.button('How can I start my company in Morocco?', on_click=promptExample1) st.sidebar.button('What are some recommended cities for starting a business in the finance sector?', on_click=promptExample2) st.sidebar.button('What is the estimated amount of money I need to start my company?', on_click=promptExample3) with st.sidebar: st.title('Disclaimer ⚠️:') st.markdown('may introduce false information') st.markdown('consult with a preofessionel advisor for more specific problems')