app.py

import os
import gradio as gr
import faiss
import numpy as np
import pickle
from sentence_transformers import SentenceTransformer
from transformers import AutoTokenizer, AutoModelForCausalLM
from preprocess import preprocess

HF_TOKEN = os.getenv("HF_TOKEN")
if not HF_TOKEN:
    raise ValueError("HF_TOKEN environment variable not set. Please configure it in Space settings.")

    
# Check if precomputed files exist and run preprocessing if needed
if not (os.path.exists("chunks.pkl") and os.path.exists("index.faiss")):
    print("Precomputed files missing. Running preprocess.py...")
    preprocess()
else:
    print("Precomputed files found. Skipping preprocessing.")

# Load precomputed chunks and FAISS index
print("Loading precomputed data...")
with open("chunks.pkl", "rb") as f:
    chunks = pickle.load(f)
index = faiss.read_index("index.faiss")

# Load embedding model (for queries only)
embedding_model = SentenceTransformer("sentence-transformers/paraphrase-multilingual-mpnet-base-v2")

# Load Jais model and tokenizer
model_name = "aubmindlab/aragpt2-base"
tokenizer = AutoTokenizer.from_pretrained(model_name, token=HF_TOKEN, trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained(model_name, token=HF_TOKEN, trust_remote_code=True)

# RAG function
def get_response(query, k=3):
    query_embedding = embedding_model.encode([query])
    distances, indices = index.search(np.array(query_embedding), k)
    retrieved_chunks = [chunks[i] for i in indices[0]]
    context = " ".join(retrieved_chunks)
    prompt = f"استنادًا إلى الوثائق التالية: {context}، أجب على السؤال: {query}"
    inputs = tokenizer(prompt, return_tensors="pt", truncation=True, max_length=512)
    outputs = model.generate(
        **inputs,
        max_new_tokens=200,
        do_sample=True,
        temperature=0.7,
        top_p=0.9
    )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split(query)[-1].strip()

# Gradio interface
with gr.Blocks(title="Dubai Legislation Chatbot") as demo:
    gr.Markdown("# Dubai Legislation Chatbot\nاسأل أي سؤال حول تشريعات دبي")
    chatbot = gr.Chatbot()
    msg = gr.Textbox(placeholder="اكتب سؤالك هنا...", rtl=True)
    clear = gr.Button("مسح")
    
    def user(user_message, history):
        return "", history + [[user_message, None]]
    
    def bot(history):
        user_message = history[-1][0]
        bot_message = get_response(user_message)
        history[-1][1] = bot_message
        return history
    
    msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
        bot, chatbot, chatbot
    )
    clear.click(lambda: None, None, chatbot, queue=False)

demo.launch()