app.py

import streamlit as st
from io import BytesIO
import ibm_watsonx_ai
import secretsload
import anton_ego_jimmy
import requests
import time
import re

from ibm_watsonx_ai.foundation_models import ModelInference
from ibm_watsonx_ai import Credentials, APIClient
from ibm_watsonx_ai.metanames import GenTextParamsMetaNames as GenParams
from ibm_watsonx_ai.metanames import GenTextReturnOptMetaNames as RetParams
from secretsload import load_stsecrets # script to load credentials from HuggingFace secrets section

credentials = load_stsecrets()

st.set_page_config(
    page_title="Jimmy le critique",
    page_icon="🍷",
    initial_sidebar_state="collapsed"
)

# Password protection
def check_password():
    def password_entered():
        if st.session_state["password"] == st.secrets["app_password"]:
            st.session_state["password_correct"] = True
            del st.session_state["password"]
        else:
            st.session_state["password_correct"] = False

    if "password_correct" not in st.session_state:
        st.markdown("\n\n")
        st.text_input("Enter the password", type="password", on_change=password_entered, key="password")
        st.divider()
        st.info("Template of Jimmy developed by Milan Mrdenovic © IBM Norway 2024")
        return False
    elif not st.session_state["password_correct"]:
        st.markdown("\n\n")
        st.text_input("Enter the password", type="password", on_change=password_entered, key="password")
        st.divider()
        st.info("Template of Jimmy developed by Milan Mrdenovic © IBM Norway 2024")
        st.error("😕 Password incorrect")
        return False
    else:
        return True

if not check_password():
    st.stop()


# Initialize session state
if 'current_page' not in st.session_state:
    st.session_state.current_page = 0

def initialize_session_state():
    if 'chat_history' not in st.session_state:
        st.session_state.chat_history = []

def setup_client():
    credentials = Credentials(
        url=st.secrets["url"],
        api_key=st.secrets["api_key"]
    )
    apo = st.secrets["api_key"]
    client = APIClient(credentials, project_id=st.secrets["project_id"])
    return client

client = setup_client()

def prepare_prompt(prompt, chat_history):
    if anton_ego_jimmy.TYPE == "chat" and chat_history:
        chats = "\n".join([f"{message['role']}: \"{message['content']}\"" for message in chat_history])
        return f"Conversation History:\n{chats}\n\nNew Message: {prompt}"
    return prompt

def apply_prompt_syntax(prompt, system_prompt, prompt_template, bake_in_prompt_syntax):
    model_family_syntax = {
        "llama3-instruct (llama-3 & 3.1) - system": """\n<|begin_of_text|><|start_header_id|>system<|end_header_id|>\n\n{system_prompt}<|eot_id|><|start_header_id|>user<|end_header_id|>\n\n{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n""",
        "llama3-instruct (llama-3 & 3.1) - user": """\n<|begin_of_text|><|start_header_id|>user<|end_header_id|>\n\n{prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n""",
        "granite-13b-chat & instruct - system": """\n<|system|>\n{system_prompt}\n<|user|>\n{prompt}\n<|assistant|>\n\n""",
        "granite-13b-chat & instruct - user": """\n<|user|>\n{prompt}\n<|assistant|>\n\n""",
        "llama2-chat - system": """\n[INST] <<SYS>>\n{system_prompt}\n<</SYS>>\n\n{prompt} [/INST]\n""",
        "llama2-chat - user": """\n[INST] {prompt} [/INST] """,
        "mistral & mixtral v2 tokenizer - system": """\n<s>[INST]System Prompt:[{system_prompt}]\n\n{prompt} [/INST]\n""",
        "mistral & mixtral v2 tokenizer - system segmented": """\n<s>[INST]System Prompt:{system_prompt}[/INST][INST]{prompt} [/INST]\n""",
        "mistral & mixtral v2 tokenizer - user": """\n<s>[INST]{prompt} [/INST]\n"""
    }
    
    if bake_in_prompt_syntax:
        template = model_family_syntax[prompt_template]
        if system_prompt:
            return template.format(system_prompt=system_prompt, prompt=prompt)
    return prompt

def generate_response(watsonx_llm, prompt_data, params):
    generated_response = watsonx_llm.generate_text_stream(prompt=prompt_data, params=params)
    for chunk in generated_response:
        yield chunk

def get_response(user_input):
    # Prepare the prompt
    prompt = prepare_prompt(user_input, st.session_state.chat_history)

    # Apply prompt syntax
    prompt_data = apply_prompt_syntax(
        prompt, 
        anton_ego_jimmy.SYSTEM_PROMPT,
        anton_ego_jimmy.PROMPT_TEMPLATE,
        anton_ego_jimmy.BAKE_IN_PROMPT_SYNTAX
    )

    watsonx_llm = ModelInference(
        api_client=client, 
        model_id=anton_ego_jimmy.SELECTED_MODEL,
        verify=anton_ego_jimmy.VERIFY
    )

    # Prepare parameters
    params = {
        GenParams.DECODING_METHOD: anton_ego_jimmy.DECODING_METHOD,
        GenParams.MAX_NEW_TOKENS: anton_ego_jimmy.MAX_NEW_TOKENS,
        GenParams.MIN_NEW_TOKENS: anton_ego_jimmy.MIN_NEW_TOKENS,
        # GenParams.LENGTH_PENALTY: anton_ego_jimmy.LENGTH_PENALTY,
        GenParams.REPETITION_PENALTY: anton_ego_jimmy.REPETITION_PENALTY,
        GenParams.STOP_SEQUENCES: anton_ego_jimmy.STOP_SEQUENCES
    }

    print(prompt_data, params)

    # Generate and stream response
    with st.chat_message("assistant", avatar="🧐"):
        stream = generate_response(watsonx_llm, prompt_data, params)
        response = st.write_stream(stream)

    # Add AI response to chat history
    st.session_state.chat_history.append({"role": "assistant", "content": response})

    return response

def main():
    initialize_session_state()
    st.title("Jimmy le critique 🧐")

    # User input
    if anton_ego_jimmy.DISPLAY_CHAT_HISTORY == 1:
        for message in st.session_state.chat_history:
            with st.chat_message(message["role"]):
                st.markdown(message["content"])
    user_input = st.chat_input("You:", key="user_input")

    if user_input:
        # Add user message to chat history
        st.session_state.chat_history.append({"role": "user", "content": user_input})
        with st.chat_message("user", avatar="🥷🏻"):
            st.markdown(user_input)

        # Get response
        get_response(user_input)

if __name__ == "__main__":
    main()