app.py

import streamlit as st
import pandas as pd
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

hf_token = os.getenv("HF_TOKEN")
token = os.getenv("HF_TOKEN")

# Load the tokenizer
model_name = "your-llama-model"  # Replace with the LLaMA model name
tokenizer = AutoTokenizer.from_pretrained(model_name)

# Load the model
model = AutoModelForCausalLM.from_pretrained(model_name)

# Apply dynamic quantization for CPU
model = torch.quantization.quantize_dynamic(
    model, {torch.nn.Linear}, dtype=torch.qint8
)

# Move model to CPU
device = torch.device("cpu")
model = model.to(device)

# Set the padding token to the end-of-sequence token
if tokenizer.pad_token is None:
    tokenizer.pad_token = tokenizer.eos_token

# Load the anomalies data
df = pd.read_csv('anomalies.csv', sep=',', decimal='.')

# Function to generate a response
def response(question):
    prompt = f"Considerando os dados: {df.to_string(index=False)}, onde a coluna 'ds' está em formato DateTime, a coluna 'real' é o valor da despesa e a coluna 'group' é o grupo da despesa. Pergunta: {question}"
    
    inputs = tokenizer(prompt, return_tensors='pt', padding='max_length', truncation=True, max_length=256).to(device)
    
    generated_ids = model.generate(
        inputs['input_ids'],
        attention_mask=inputs['attention_mask'],
        max_length=inputs['input_ids'].shape[1] + 50,
        temperature=0.7,
        top_p=0.9,
        no_repeat_ngram_size=2,
        num_beams=3,
    )
    
    generated_text = tokenizer.decode(generated_ids[0], skip_special_tokens=True)
    final_response = generated_text.split("Resposta:")[-1].split(".")[0] + "."
    
    return final_response

# Streamlit interface
st.markdown("""
<div style='display: flex; align-items: center;'>
    <div style='width: 40px; height: 40px; background-color: green; border-radius: 50%; margin-right: 5px;'></div>
    <div style='width: 40px; height: 40px; background-color: red; border-radius: 50%; margin-right: 5px;'></div>
    <div style='width: 40px; height: 40px; background-color: yellow; border-radius: 50%; margin-right: 5px;'></div>
    <span style='font-size: 40px; font-weight: bold;'>Chatbot do Tesouro RS</span>
</div>
""", unsafe_allow_html=True)

# Chat history
if 'history' not in st.session_state:
    st.session_state['history'] = []

# Input box for user question
user_question = st.text_input("Escreva sua questão aqui:", "")

if user_question:
    # Add person emoji when typing question
    st.session_state['history'].append(('👤', user_question))
    st.markdown(f"**👤 {user_question}**")
    
    # Generate the response
    bot_response = response(user_question)
    
    # Add robot emoji when generating response and align to the right
    st.session_state['history'].append(('🤖', bot_response))
    st.markdown(f"<div style='text-align: right'>**🤖 {bot_response}**</div>", unsafe_allow_html=True)

# Clear history button
if st.button("Limpar"):
    st.session_state['history'] = []

# Display chat history
for sender, message in st.session_state['history']:
    if sender == '👤':
        st.markdown(f"**👤 {message}**")
    elif sender == '🤖':
        st.markdown(f"<div style='text-align: right'>**🤖 {message}**</div>", unsafe_allow_html=True)