Pisi_u_stilu_FT.py

# program za pisanje u stilu neke osobe, uzima stil i temu iz Pinecone indexa

# uvoze se biblioteke
import os
import streamlit as st
import pinecone
from langchain.vectorstores.pinecone import Pinecone
from langchain.embeddings.openai import OpenAIEmbeddings
from langchain.vectorstores import Pinecone
from langchain.chat_models import ChatOpenAI
from langchain.chains import LLMChain
from langchain import LLMChain
from langchain.prompts.chat import (
    SystemMessagePromptTemplate,
    HumanMessagePromptTemplate,
    ChatPromptTemplate,
)
from html2docx import html2docx
from myfunc.mojafunkcija import st_style, positive_login, open_file
import markdown
from langchain.utilities import GoogleSerperAPIWrapper
import pdfkit


# these are the environment variables that need to be set for LangSmith to work


version = "09.10.23. - 3"


def main():
    GOOGLE_API_KEY = os.environ.get("GOOGLE_API_KEY")
    GOOGLE_CSE_ID = os.environ.get("GOOGLE_CSE_ID")
    if "SERPER_API_KEY" not in st.session_state:
        st.session_state.serper_api_key = os.environ.get("SERPER_API_KEY")

    # Retrieving API keys from env
    openai_api_key = os.environ.get("OPENAI_API_KEY")
    # Initialize Pinecone
    pinecone.init(
        api_key=os.environ["PINECONE_API_KEY"],
        environment=os.environ["PINECONE_API_ENV"],
    )
    # Initialize OpenAI embeddings
    embeddings = OpenAIEmbeddings()
    search = GoogleSerperAPIWrapper()
    # Initialize OpenAI embeddings and LLM and all variables

    if "model" not in st.session_state:
        st.session_state.model = ""
    if "temp" not in st.session_state:
        st.session_state.temp = 1.0
    if "text" not in st.session_state:
        st.session_state.text = "text"
    if "namespace" not in st.session_state:
        st.session_state.namespace = "koder"
    if "index_name" not in st.session_state:
        st.session_state.index_name = "embedings1"
    if "odgovor" not in st.session_state:
        st.session_state.odgovor = ""
    if "tematika" not in st.session_state:
        st.session_state.tematika = ""
    if "thold" not in st.session_state:
        st.session_state.thold = 0.5
    if "stil" not in st.session_state:
        st.session_state.stil = ""

    # Izbor stila i teme
    st.markdown(
        f"<p style='font-size: 10px; color: grey;'>{version}</p>",
        unsafe_allow_html=True,
    )
    st.subheader("Pišite u stilu osoba koje imaju sopstvene Fine-Tunned modele 🏙️")
    with st.expander("Pročitajte uputstvo 🧝"):
        st.caption(
            """
                   FT se odnosi na Fine-Tuning, tj. prilagođavanje aplikacije nekoj specifičnoj primeni (iliti specijalizacija)
                   - u našem slučaju se aplikacija prilagođava nečijem stilu pisanja (npr. od Miljana).\n
                   Promptove možete naći na Public-u - folder AI Dev.
                   """
        )
        st.image(
            "https://test.georgemposi.com/wp-content/uploads/2023/09/PisiUStilu1.png"
        )
        st.caption(
            """\n
                   1.	Parametri za podešavanje rada aplikacije - opisani su u levom meniju, a i intuitivni su.\n
                   2.	Uploadovanje ili direktno kucanje teksta/teme o kojoj biste da pišete.\n
                   Ono što uploadujete će se prikazati u tekstualnom polju ispod - to polje je ono što aplikacija gleda kada se izvršava.\n
                   3.	Ovde je obrnuto u odnosu na Multi Tool Chatbot - prvo se unosi komentar, pa se onda ocenjuje (slika ispod).
                   """
        )
        st.image(
            "https://test.georgemposi.com/wp-content/uploads/2023/09/PisiUStilu2.png"
        )
        st.caption(
            """\n
                   1.	Generisani tekst i opcije za skidanje teksta na računar u različitim oblicima.\n
                   2.	Komentar koji ste upisali, pa kliknuli Enter ili strelicu u uglu polja za komentarsanje.\n
                   3.	Ocenjivanje od 1 do 5.\n
                   4.	Polje za unos komentara je sada zaključano - mora refresh stranice da bi se aplikacija opet koristila.\n
                   """
        )

    st.caption(
        """
               Ova aplikacija omogućava generisanje teksta na određenu temu i da se koristi kao osnova za pisanje teksta u stilu
               odabrane osobe.\n Koristi se Pinecone indeks za pronalaženje teksta na određenu temu.
               Ukoliko ne pronađe odgovarajući tekst, potražiće odgovor na internetu.
               """
    )
    with st.sidebar:
        st.session_state.namespace = st.selectbox(
            "Odaberite oblast", ("koder", "positive")
        )

        ft_model = st.selectbox(
            "Odaberite model", ("Dragan Simic", "Miljan Radanovic", "Pera Lozac")
        )

        if ft_model == "Dragan Simic":
            st.session_state.model = (
                "ft:gpt-3.5-turbo-0613:positive-doo:dragan-simic:7rLzG9Cp"
            )
            st.session_state.stil = "Dragan Simic is an IT expert. He writes in a long sentences in overly polite manner. He always writes in the Serbian language"
        elif ft_model == "Miljan Radanovic":
            st.session_state.model = (
                "ft:gpt-3.5-turbo-0613:positive-doo:miljan:7rIDKWid"
            )
            st.session_state.stil = "Miljan Radanovic is an IT expert. He writes in a long sentences and offten mixes complex and everyday terms in the same sentence. He always writes in the Serbian language"
        elif ft_model == "Pera Lozac":
            st.session_state.model = (
                "ft:gpt-3.5-turbo-0613:positive-doo:pera-lozac:7rKBrShJ"
            )
            st.session_state.stil = "Pera Lozac knows the answers, but he writes in a short sentences in a style of disfluent person and use verbal crutches"

        st.session_state.temp = st.slider(
            "Set temperature (0=strict, 1=creative)", 0.0, 2.0, step=0.1, value=1.0
        )
        st.caption("Temperatura za stil treba de je što bliže 1.0")
        st.session_state.thold = st.slider(
            "Set relevance (0=any, 1=strict)", 0.0, 1.0, step=0.1, value=0.5
        )
        st.caption(
            "Relevantnost za temu određuje koji dokmenti će se korsititi iz indeksa. Ako je vrednost 0.0 onda se koriste svi dokumenti, a za 1.0 samo oni koji su najrelevantniji."
        )

    # define model, vestorstore and retriever
    llm = ChatOpenAI(
        model_name=st.session_state.model,
        temperature=st.session_state.temp,
        openai_api_key=openai_api_key,
    )
    vectorstore = Pinecone.from_existing_index(
        st.session_state.index_name,
        embeddings,
        st.session_state.text,
        namespace=st.session_state.namespace,
    )

    # Prompt template - Loading text from the file
    prompt_file = st.file_uploader(
        "Izaberite početni prompt koji možete editovati ili pišite prompt od početka za definisanje vašeg zahteva",
        key="upload_prompt",
        type="txt",
    )
    prompt_t = ""
    if prompt_file is not None:
        prompt_t = prompt_file.getvalue().decode("utf-8")
    else:
        prompt_t = " "

    # Prompt
    with st.form(key="stilovi", clear_on_submit=False):
        zahtev = st.text_area(
            "Opišite temu, iz oblasti Positive, ili opšte teme. Objasnite i formu željenog teksta: ",
            prompt_t,
            key="prompt_prva",
            height=150,
        )
        submit_button = st.form_submit_button(label="Submit")
        st.session_state.tematika = vectorstore.similarity_search_with_score(
            zahtev, k=3
        )
    # pocinje obrada, prvo se pronalazi tematika, zatim stil i na kraju se generise odgovor
    if submit_button:
        with st.spinner("Obrađujem temu..."):
            broj = 1
            doclist = []
            uk_teme = ""

            # Iterate through the documents in st.session_state.tematika with enumerate
            for broj, (doc, score) in enumerate(st.session_state.tematika, start=1):
                # Check if the similarity score is greater than st.session_state.thold
                if score > st.session_state.thold:
                    # Append the page content to the selected_docs list
                    doclist.append(doc.page_content)
                    st.info(
                        f"Score sličnosti za dokument broj {broj} je: {round(score, 2)}"
                    )
                    # Now, selected_docs contains the page content of documents with a score greater than st.session_state.thold
            uk_teme = doclist
            # ako ne pronadje temu u indexu, trazi na internetu
            if len(doclist) == 0:
                st.info(
                    "Nisam u mogućnosti da pronađem odgovor u indeksu. Pretražujem internet..."
                )
                uk_teme = search.results(zahtev)
            st.info(
                f"Za relevantnost veću od {st.session_state.thold} broj pronađenih dokumenata je {len(doclist)} "
            )
            st.info(
                f"Korišćen je model '{ft_model}' - temperatura je {st.session_state.temp}"
            )

            # Read prompt template from the file
            sve_zajedno = open_file("prompt_FT.txt")
            system_message_prompt = SystemMessagePromptTemplate.from_template(
                st.session_state.stil
            )
            system_message = system_message_prompt.format()
            human_message_prompt = HumanMessagePromptTemplate.from_template(sve_zajedno)
            human_message = human_message_prompt.format(
                zahtev=zahtev, uk_teme=uk_teme, ft_model=ft_model
            )
            prompt = ChatPromptTemplate(messages=[system_message, human_message])

            # Create LLM chain with chatbot prompt
            chain = LLMChain(llm=llm, prompt=prompt)

            with st.expander("Model i Prompt", expanded=False):
                st.write(
                    f"Korišćen je prompt: {prompt.messages[0].content} ->  {prompt.messages[1].content} - >"
                )
            # Run chain to get chatbot's answer
            with st.spinner("Pišem tekst..."):
                try:
                    st.session_state.odgovor = chain.run(prompt=prompt)
                except Exception as e:
                    st.warning(
                        f"Nisam u mogućnosti da završim tekst. Ovo je opis greške:\n {e}"
                    )

    # Izrada verzija tekstova za fajlove formnata po izboru
    # html to docx
    if st.session_state.odgovor != "":
        with st.expander("FINALNI TEKST", expanded=True):
            st.markdown(st.session_state.odgovor)
        html = markdown.markdown(st.session_state.odgovor)
        buf = html2docx(html, title="Zapisnik")

        options = {
            "encoding": "UTF-8",  # Set the encoding to UTF-8
            "no-outline": None,
            "quiet": "",
        }
        try:
            pdf_data = pdfkit.from_string(html, cover_first=False, options=options)
            st.download_button(
                label="Download TekstuStilu.pdf",
                data=pdf_data,
                file_name="TekstuStilu.pdf",
                mime="application/octet-stream",
            )
        except:
            st.write(
                "Za pdf fajl restartujte app za 5 minuta. Osvezavanje aplikacije je u toku"
            )
        st.download_button(
            "Download TekstuStilu.txt",
            st.session_state.odgovor,
            file_name="TekstuStilu.txt",
        )

        st.download_button(
            label="Download TekstuStilu.docx",
            data=buf.getvalue(),
            file_name="TekstuStilu.docx",
            mime="docx",
        )

    # if prompt := st.chat_input(placeholder="Unesite komentare na rad programa."):
    #     st.session_state["user_feedback"] = prompt
    #     st.chat_input(placeholder="Feedback je sačuvan!", disabled=True)
    #     st.session_state.feedback = None
    #     st.session_state.feedback_update = None
    #     run_collector = RunCollectorCallbackHandler()

    #     prompt = ChatPromptTemplate.from_messages([("system", "Hi"), ("human", "Hi")])
    #     llm = ChatOpenAI(temperature=0.7)
    #     chain = LLMChain(prompt=prompt, llm=llm)

    #     x = chain.invoke(
    #         {"input": "Hi."},
    #         config=RunnableConfig(
    #             callbacks=[run_collector],
    #             tags=["Streamlit Chat"],
    #         ),
    #     )["text"]

    #     run = run_collector.traced_runs[0]
    #     run_collector.traced_runs = []
    #     st.session_state.run_id = run.id
    #     wait_for_all_tracers()
    #     try:
    #         client.share_run(run.id)
    #     except ValueError:
    #         st.write("...")

    # if st.session_state.get("run_id"):
    #     with st.chat_message("assistant", avatar="🤖"):
    #         message_placeholder = st.empty()
    #         message_placeholder.markdown(
    #             ":rainbow[Samo još ocenite od 1 do 5 dobijene rezultate.]"
    #         )
    #     feedback = streamlit_feedback(
    #         feedback_type="faces", key=f"feedback_{st.session_state.run_id}"
    #     )
    #     scores = {"😞": 1, "🙁": 2, "😐": 3, "🙂": 4, "😀": 5}
    #     if feedback:
    #         score = scores[feedback["score"]]
    #         feedback = client.create_feedback(
    #             st.session_state.run_id,
    #             "ocena",
    #             score=score,
    #             comment=st.session_state["user_feedback"],
    #         )
    #         st.session_state.feedback = {
    #             "feedback_id": str(feedback.id),
    #             "score": score,
    #         }

    # if st.session_state.get("feedback"):
    #     feedback = st.session_state.get("feedback")
    #     x = ["🎭", "🐯", "👺", "👻", "😸", "🤓", "🤡", "🦄", "🧟‍♀️", "☘️"]
    #     st.write(
    #         f"{x[randint(0, len(x) - 1)]} Ova aplikacija NE radi iterativno - mora refresh stranice!"
    #     )
    #     st.chat_input(placeholder="To je to - hvala puno!", disabled=True)


# Login
st_style()
# Koristi se samo za deploy na streamlit.io
deployment_environment = os.environ.get("DEPLOYMENT_ENVIRONMENT")

if deployment_environment == "Streamlit":
    name, authentication_status, username = positive_login(main, " ")
else:
    if __name__ == "__main__":
        main()