app.py

import logging
import os
import time

import streamlit as st
import torch
import sys

from llama_index.core import VectorStoreIndex, SimpleDirectoryReader, Settings, StorageContext, load_index_from_storage
from llama_index.core.chat_engine.types import ChatMode
from llama_index.embeddings.huggingface import HuggingFaceEmbedding
from llama_index.llms.openai_like import OpenAILike

PAGE_TITLE="Votre expert Scrum"
CHAT_TITLE="Posez-moi une question sur le guide Scrum 2020 (anglais ou français)"
SYSTEM_PROMPT="Use the context information provided to assist the user. Mention the origins of the informations at the bottom of the response (file and page)."
#EMBEDDING_MODEL="sentence-transformers/paraphrase-MiniLM-L6-v2" # Fast embedding model
EMBEDDING_MODEL="BAAI/bge-small-en-v1.5"
#EMBEDDING_MODEL="BAAI/bge-m3" # Multilingual large model
#LLM_MODEL="DeepSeek-R1-Distill-Llama-70B" # Available models on : https://chatapi.akash.network/documentation#models
LLM_MODEL="DeepSeek-R1-Distill-Qwen-32B"
NB_DOC_CHUNKS_TO_SEND=5
MAX_NB_TOKENS_IN_RESPONSE=1500
TEMPERATURE=0.5 # The closer to 1, the less deterministic and the more creative

API_BASE_URL="https://chatapi.akash.network/api/v1" # Changing this requires to adapt the custom_llm initialization

# Ajuster le chemin de torch.classes pour éviter le conflit
torch.classes.__path__ = []

st.set_page_config(page_title=PAGE_TITLE, layout="centered", initial_sidebar_state="auto", menu_items=None)
st.title(PAGE_TITLE)

custom_llm = OpenAILike(model=LLM_MODEL, api_base=API_BASE_URL, api_key=st.secrets["openai_key"], max_tokens=MAX_NB_TOKENS_IN_RESPONSE, temperature=TEMPERATURE)
Settings.embed_model = HuggingFaceEmbedding(model_name=EMBEDDING_MODEL)
Settings.llm=custom_llm

logging.basicConfig(stream=sys.stdout, level=logging.DEBUG)
logging.getLogger().addHandler(logging.StreamHandler(stream=sys.stdout))

# Load and index data
@st.cache_resource
def load_data():
    persist_dir = "./storage"
    if not os.path.exists(persist_dir):
        documents = SimpleDirectoryReader(input_dir="./data").load_data()
        document_index = VectorStoreIndex.from_documents(documents)
        document_index.storage_context.persist(persist_dir=persist_dir)
    else:
        storage_context = StorageContext.from_defaults(persist_dir=persist_dir)
        document_index = load_index_from_storage(storage_context)
    return document_index

start_time = time.time()

index = load_data()

end_time = time.time()
print(f"Time taken for loading embeddings: {end_time - start_time:.4f} seconds")
start_time = time.time()

if "messages" not in st.session_state.keys():  # Initialize the chat messages history
    st.session_state.messages = [
        {
            "role": "assistant",
            "content": CHAT_TITLE,
        }
    ]

if "chat_engine" not in st.session_state.keys():  # Initialize the chat engine
    st.session_state.chat_engine = index.as_chat_engine(chat_mode=ChatMode.CONTEXT, system_prompt=SYSTEM_PROMPT, similarity_top_k=NB_DOC_CHUNKS_TO_SEND, verbose=True, streaming=True)

if prompt := st.chat_input("Posez votre question"):  # Prompt for user input and save to chat history
    st.session_state.messages.append({"role": "user", "content": prompt})

for message in st.session_state.messages:  # Write message history to UI
    with st.chat_message(message["role"]):
        st.write(message["content"])

# If last message is not from assistant, generate a new response
if st.session_state.messages[-1]["role"] != "assistant":
    with st.chat_message("assistant"):

        start_time = time.time()

        response_stream = st.session_state.chat_engine.stream_chat(prompt)
        st.write_stream(response_stream.response_gen)
        message = {"role": "assistant", "content": response_stream.response}
        # Add response to message history
        st.session_state.messages.append(message)

        end_time = time.time()
        print(f"Time taken for getting response: {end_time - start_time:.4f} seconds")
        start_time = time.time()