src/app.py

"""Interface de chat com modelo de linguagem."""

import streamlit as st
import base64
from pathlib import Path
from backend import load_model, ChatModel
from config import get_model_options
from send_to_hub import send_message_buckt
import uuid
import logging

logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')

st.set_page_config(
    layout="wide",
    page_title="Small LLM - Chat",
    initial_sidebar_state="collapsed"
)

def set_new_message_uuid():
    st.session_state.message_uuid = uuid.uuid4()
    logging.info(f'[NEW MESSAGE UUID] {st.session_state.message_uuid}')

def save_messages():
    if "chat_model" not in st.session_state or not st.session_state.chat_model.conversation.messages:
        return
        
    messages = [ {'role':m.role, 'content':m.content} for m in st.session_state.chat_model.conversation.messages]
    message_uuid = st.session_state.message_uuid
    
    new_data = {}
    if "model_info" in st.session_state:
        for k,v in st.session_state.model_info.items():
            new_data[str(k)] = str(v)
    
    new_data['model_name'] = str(st.session_state.get('model_name', 'unknown'))
    
    send_message_buckt(messages, message_uuid, new_data, REPO_ID = "CEIA-POSITIVO2/smol_chat")


# Caminho da logo (relativo à raiz do projeto)
PROJECT_ROOT = Path(__file__).parent.parent
LOGO_PATH = PROJECT_ROOT / "positivo-logo.png"

# Header com logo e título usando HTML/CSS para melhor controle
logo_html = ""
if LOGO_PATH.exists():
    try:
        with open(LOGO_PATH, "rb") as img_file:
            img_base64 = base64.b64encode(img_file.read()).decode()
        logo_html = f'<img src="data:image/png;base64,{img_base64}" />'
    except Exception:
        logo_html = ""

st.markdown(f"""
<style>
.logo-header {{
    display: flex;
    align-items: center;
    gap: 20px;
    margin-bottom: 0.5rem;
}}
.logo-header img {{
    width: 90px;
    height: 90px;
    object-fit: contain;
    flex-shrink: 0;
}}
</style>
<div class="logo-header">
    {logo_html}
    <h1 style="margin: 0; padding: 0; display: inline-block;">Small LLM - Chat</h1>
</div>
""", unsafe_allow_html=True)

def handle_model_load_error(model_name: str, error_msg: str):
    """Trata erros de carregamento de modelo."""
    # Nota: Assumindo que GATED_MODELS está definido em config ou globalmente, 
    # caso contrário essa verificação precisaria ser ajustada.
    st.error(f"❌ Erro ao carregar modelo {model_name}: {error_msg}")

# ============================================================================
# LÓGICA DE CARREGAMENTO AUTOMÁTICO (Substitui a Sidebar)
# ============================================================================

if "chat_model" not in st.session_state:
    with st.spinner("Inicializando o sistema e carregando o modelo..."):
        try:
            # 1. Obter opções de modelo
            model_options = get_model_options() # tuple(model_id, config)
            
            if not model_options:
                st.error("Nenhum modelo encontrado nas configurações.")
                st.stop()

            # 2. Selecionar automaticamente o PRIMEIRO modelo da lista
            selected_label, selected_model_config = model_options[2]
            logging.info(f'[MODEL] Loading {selected_label} {selected_model_config}')
            
            # Preparar configuração
            model_name = selected_model_config.pop('model_id')
            selected_model_config['dtype'] = 'bfloat16'
            
            print(f'Carregando modelo automático: {model_name}')

            # 3. Carregar o modelo
            pipeline, model_info = load_model(
                model_name,
                **selected_model_config
            )
            
            # 4. Salvar na sessão
            chat_model = ChatModel(pipeline)
            st.session_state.chat_model = chat_model
            st.session_state.model_info = selected_model_config
            st.session_state.model_name = model_name
            set_new_message_uuid()
            
            # Força o rerun para atualizar a interface sem o spinner
            st.rerun()
            
        except Exception as e:
            handle_model_load_error(model_name if 'model_name' in locals() else "Unknown", str(e))
            st.stop()

# ============================================================================
# INTERFACE DE CHAT
# ============================================================================

chat_model = st.session_state.chat_model

if "messages" not in st.session_state:
    st.session_state.messages = []

# Sincroniza mensagens do objeto chat_model com a session_state
if len(chat_model.conversation.messages) != len(st.session_state.messages):
    st.session_state.messages = [ {"role": msg.role, "content": msg.content} for msg in chat_model.conversation.messages ]

chat_placeholder = st.empty()

with chat_placeholder.container():
    for message in st.session_state.messages:
        role = message["role"]
        content = message["content"]
        
        if role == "system":
            continue
       
        with st.chat_message(role):
            st.markdown(content)

if user_input := st.chat_input("Digite sua mensagem..."):
        # Limpa o historico
        st.session_state.messages = []
        chat_model.conversation.clear()
        chat_placeholder.empty()
        set_new_message_uuid()
    
        chat_model.add_user_message(user_input)
        save_messages()
        st.session_state.messages.append({"role": "user", "content": user_input})
    
        with st.chat_message("user"):
            st.markdown(user_input)
    
        with st.chat_message("assistant"):
            response_placeholder = st.empty()
            full_response = ""
            try:
                for token in chat_model.generate_streaming(max_new_tokens=4096):
                    full_response += token
                    response_placeholder.markdown(full_response)
            
                chat_model.add_assistant_message(full_response)
                st.session_state.messages.append({"role": "assistant", "content": full_response})
                save_messages()
            
            except Exception as e:
                error_msg = f"Erro na geração: {str(e)}"
                st.error(error_msg)
                st.session_state.messages.append({"role": "assistant", "content": error_msg})