app.py

import os
from dotenv import load_dotenv
import gradio as gr
from huggingface_hub import InferenceClient, login
from langchain_huggingface import HuggingFaceEmbeddings
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_community.vectorstores import FAISS
from langchain_community.document_loaders import PyPDFLoader
from langchain.schema import Document

load_dotenv()
hftoken = os.getenv("HFTOKEN")
prompt_systeme = os.getenv("PROMPT_SYSTEM")
questions = os.getenv("QUESTIONS").split(",")  


login(token=hftoken)
client = InferenceClient(model="meta-llama/Llama-3.3-70B-Instruct", token=hftoken)
embeddings = HuggingFaceEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2")

vector_store = None

def extract_text_from_file(file_path):
    """Extrait le texte d'un fichier PDF ou TXT."""
    try:
        ext = os.path.splitext(file_path)[1].lower()
        if ext == ".pdf":
            loader = PyPDFLoader(file_path)
            pages = loader.load()
            return [Document(page_content=page.page_content) for page in pages]
        elif ext == ".txt":
            with open(file_path, "r", encoding="utf-8") as file:
                return [Document(page_content=file.read())]
        else:
            return None
    except Exception as e:
        print(f"Erreur extraction : {e}")
        return None

def embed_documents(file):
    """Charge un document dans FAISS pour la recherche vectorielle."""
    global vector_store
    docs = extract_text_from_file(file.name)
    if not docs:
        return "❌ Erreur lors du chargement du fichier."
    
    text_splitter = RecursiveCharacterTextSplitter(chunk_size=500, chunk_overlap=100)
    documents = text_splitter.split_documents(docs)

    if documents:
        vector_store = FAISS.from_documents(documents, embeddings)
        return "📁 Document chargé et indexé ! Posez-moi une question."
    else:
        return "❌ Aucun texte valide extrait du fichier."

def query_faiss(query):
    """Effectue une recherche dans FAISS et génère une réponse."""
    global vector_store
    if vector_store is None:
        return None
    
    retriever = vector_store.as_retriever(search_kwargs={"k": 3})
    results = retriever.get_relevant_documents(query)

    if results:
        context = "\n".join([doc.page_content for doc in results])
        return context
    return None

def chatbot_response(message, history, system_message, max_tokens, temperature, top_p, file=None):
    """Gère les réponses du chatbot avec ou sans document."""
    global vector_store
    if file:
        status = embed_documents(file)
        if "❌" in status:
            yield status
            return
        yield status

    context = query_faiss(message) if vector_store else None
    messages = [{"role": "system", "content": system_message}]
    
    for val in history:
        if val[0]: messages.append({"role": "user", "content": val[0]})
        if val[1]: messages.append({"role": "assistant", "content": val[1]})

    if context:
        messages.append({"role": "user", "content": f"Contexte : {context}\nQuestion : {message}"})
    else:
        messages.append({"role": "user", "content": message})

    response = ""
    try:
        response_stream = client.chat_completion(
            messages,
            max_tokens=max_tokens,
            stream=True,
            temperature=temperature,
            top_p=top_p,
        )

        for msg in response_stream:
            if msg.choices and msg.choices[0].delta and msg.choices[0].delta.content:
                token = msg.choices[0].delta.content
                response += token
                yield response

        if not response.strip():
            yield "Je ne sais pas. Peux-tu reformuler ?"
    except Exception as e:
        print(f"Erreur réponse : {e}")
        yield "❌ Une erreur est survenue lors de la génération de la réponse."

def handle_question(selected_question):
    """Gère la sélection d'une question pré-remplie."""
    return selected_question.strip()  



with gr.Blocks(theme=gr.themes.Soft()) as demo:
    with gr.Row():
        gr.Image("logo-gaia.png", width=300, height=300, show_label=False, show_download_button=False)

    with gr.Row():
        gr.Markdown("<h1 style='text-align: center;'>📚 Chatbot GAIA</h1>")

    with gr.Row():
        gr.Markdown("<p style='text-align: center;'>Exemple : Allez sur <a href='https://www.societe.ninja/' target='_blank'>société.ninja</a> avec votre numéro de SIREN et utilisez un PDF.</p>")

    with gr.Row():
        with gr.Column():
            gr.Markdown("## ⚙️ Paramètres")
            with gr.Accordion(" Paramètres avancés", open=False):
                system_message = gr.Textbox(value=prompt_systeme, label="Message système")
                max_tokens = gr.Slider(1, 2048, value=800, step=1, label="Max tokens")  
                temperature = gr.Slider(0.1, 4.0, value=0.3, step=0.1, label="Température")
                top_p = gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p")

            gr.Markdown("## 📂 Télécharger un fichier")
            file_upload = gr.File(label="Téléchargez un PDF ou TXT", file_types=[".pdf", ".txt"], type="filepath")
            file_upload.change(embed_documents, inputs=file_upload, outputs=[])
            gr.Markdown("## ❓ Questions pré-remplies")
            question_buttons = []
            for question in questions:
                button = gr.Button(value=question.strip())
                question_buttons.append(button)

            selected_question_output = gr.Textbox(label="Question sélectionnée", interactive=False)

            for button in question_buttons:
                button.click(handle_question, inputs=button, outputs=selected_question_output)

        with gr.Column():
            gr.Markdown("## 💬 Chat")
            chatbot = gr.ChatInterface(
                chatbot_response,
                additional_inputs=[system_message, max_tokens, temperature, top_p, file_upload],
            )

if __name__ == "__main__":
    demo.launch(share=True)