import openai import streamlit as st from langchain.llms import OpenAI from langchain.chat_models import ChatOpenAI from langchain.embeddings import HuggingFaceEmbeddings from langchain.chains import RetrievalQA from langchain.prompts.prompt import PromptTemplate from langchain.vectorstores import FAISS import re import time # import e5-large-v2 embedding model model_name = "intfloat/e5-large-v2" model_kwargs = {'device': 'cpu'} encode_kwargs = {'normalize_embeddings': False} embeddings = HuggingFaceEmbeddings( model_name=model_name, model_kwargs=model_kwargs, encode_kwargs=encode_kwargs ) # load IPCC database db = FAISS.load_local("IPCC_index_e5_1000_all", embeddings) question1 = 'Why does temperature increase?' question2 = 'What evidence we have of climate change?' question3 = 'What is the link between health and climate change?' def click_button(button_text): if prompt := button_text: #if prompt := st.text_input(label="Your quesiton:",value=st.session_state.button_text if 'button_text' in st.session_state else 'Text your question'): if not openai_api_key: st.info("Please add your OpenAI API key to continue.") st.stop() st.session_state.messages.append({"role": "user", "content": prompt}) st.chat_message("user").write(prompt) with st.spinner("Thinking..."): result = generate_response(prompt) result_r = result["result"] index = result_r.find("Highlight:") # Display assistant response in chat message container with st.chat_message("assistant"): message_placeholder = st.empty() full_response = "" assistant_response = result_r # Simulate stream of response with milliseconds delay for chunk in assistant_response.split(): full_response += chunk + " " time.sleep(0.05) # Add a blinking cursor to simulate typing message_placeholder.write(full_response + "โ–Œ") message_placeholder.write(result_r) # Add assistant response to chat history st.session_state.messages.append({"role": "assistant", "content": result_r}) def generate_response(input_text): docs = db.similarity_search(input_text,k=5) json1 = docs[0].metadata json2 = docs[1].metadata json3 = docs[2].metadata json4 = docs[3].metadata json5 = docs[4].metadata #st.write({"source1":json1["source"], "source2":json2["source"],"source3":json3["source"]}) climate_TEMPLATE = """\ You are a professor in climate change, tasked with answering any question \ about climate change. Take a deep breath and think step by step. {question} Generate a comprehensive and informative answer and three next questions to the general audience of 100 words or less for the \ given question based solely on the provided search results (hyperlink and source). You must \ only use information from the provided search results. Use an unbiased and \ journalistic tone. Combine search results together into a coherent answer. Do not \ repeat text. Only use \ relevant results that answer the question accurately. list these sources at the end of your answer \ in a section named "source". After the "source" section, makre sure provide three next questions in the section of predicted \ \ Format your answer in markdown format If there is nothing in the context relevant to the question at hand, just say "Hmm, \ I'm not sure." Don't try to make up an answer. Anything between the following `context` html blocks is retrieved from a knowledge \ bank, not part of the conversation with the user. {context} Anything between the following `sources` html blocks is the source and hyperlink you should use and list them into a source section\ [{source1} page {page1}](https://www.ipcc.ch/report/ar6/{wg1}/downloads/report/{source1}.pdf#page={page1}) [{source2} page {page2}](https://www.ipcc.ch/report/ar6/{wg2}/downloads/report/{source2}.pdf#page={page2}) [{source3} page {page3}](https://www.ipcc.ch/report/ar6/{wg3}/downloads/report/{source3}.pdf#page={page3}) [{source4} page {page4}](https://www.ipcc.ch/report/ar6/{wg4}/downloads/report/{source4}.pdf#page={page4}) [{source5} page {page5}](https://www.ipcc.ch/report/ar6/{wg5}/downloads/report/{source5}.pdf#page={page5}) REMEMBER: If there is no relevant information within the context, just say "Hmm, I'm \ not sure." Don't try to make up an answer. Anything between the preceding 'context' \ html blocks is retrieved from a knowledge bank, not part of the conversation with the \ user.\ """ climate_PROMPT = PromptTemplate(input_variables=["question", "context"], partial_variables={"source1":json1["source"], "source2":json2["source"], "source3":json3["source"],"source4":json4["source"],"source5":json5["source"],"page1":json1["page"], "page2":json2["page"],"page3":json3["page"],"page4":json4["page"],"page5":json5["page"],"wg1":json1["wg"], "wg2":json2["wg"],"wg3":json3["wg"],"wg4":json4["wg"],"wg5":json5["wg"]}, template=climate_TEMPLATE, ) #climate_PROMPT.partial(source = docs[0].metadata) llm = ChatOpenAI( model_name="gpt-3.5-turbo-16k", temperature=0.05, max_tokens=2500, openai_api_key=openai_api_key ) # Define retriever retriever = db.as_retriever(search_kwargs={"k": 5}) qa_chain = RetrievalQA.from_chain_type(llm, retriever=retriever, chain_type="stuff", #"stuff", "map_reduce","refine", "map_rerank" return_source_documents=True, verbose=True, chain_type_kwargs={"prompt": climate_PROMPT} ) return qa_chain({'query': input_text}) with st.sidebar: openai_api_key = st.text_input("OpenAI API Key", key="chatbot_api_key", type="password") "[Get an OpenAI API key](https://platform.openai.com/account/api-keys)" st.markdown("## ๐ŸŒ Welcome to ClimateChat! ๐ŸŒ") st.markdown("ClimateChat harnesses the latest [IPCC reports](https://www.ipcc.ch/report/sixth-assessment-report-cycle/) and the power of Large Language Models to answer your questions about climate change. When you interact with ClimateChat not only will you receive clear answers, but each response is coupled with sources and hyperlinks for further exploration and verification.\ Our objective is to make climate change information accessible, understandable, and actionable for everyone, everywhere.") st.title("๐Ÿ’ฌ๐ŸŒ๐ŸŒก๏ธClimateChat") st.caption("๐Ÿ’ฌ A Climate Change chatbot powered by OpenAI LLM and IPCC documents") #col1, col2, = st.columns(2) if "messages" not in st.session_state: st.session_state["messages"] = [{"role": "assistant", "content": "Any question about the climate change? Here are some examples:"}] for msg in st.session_state.messages: st.chat_message(msg["role"]).write(msg["content"]) if prompt := st.chat_input(): #if prompt := st.text_input(label="Your quesiton:",value=st.session_state.button_text if 'button_text' in st.session_state else 'Text your question'): if not openai_api_key: st.info("Please add your OpenAI API key to continue.") st.stop() st.session_state.messages.append({"role": "user", "content": prompt}) st.chat_message("user").write(prompt) with st.chat_message("assistant"): with st.spinner("thinking..."): result = generate_response(prompt) result_r = result["result"] index = result_r.find("Highlight:") message_placeholder = st.empty() full_response = "" assistant_response = result_r # Simulate stream of response with milliseconds delay for chunk in assistant_response.split(): full_response += chunk + " " time.sleep(0.05) # Add a blinking cursor to simulate typing message_placeholder.write(full_response + "โ–Œ") message_placeholder.write(result_r) if re.search(r'Next Questions?:', result_r, flags=re.IGNORECASE): questions_text = re.split(r'Next Questions?:', result_r, flags=re.IGNORECASE)[-1] pattern = r'\d+\.\s*([^?]+\?)' # Use re.findall to find all matches of the pattern in the questions_text matches = re.findall(pattern, questions_text) question1 = matches[0].strip() question2 = matches[1].strip() question3 = matches[2].strip() # Add assistant response to chat history st.session_state.messages.append({"role": "assistant", "content": result_r}) button_col1, button_col2, button_col3= st.columns([5, 5, 5]) st.code(question1,language =None) st.code(question2,language=None) st.code(question3,language=None)