app.py

import os
from dotenv import find_dotenv, load_dotenv
import streamlit as st
from typing import Generator
from groq import Groq

# Cargar variables de entorno
_ = load_dotenv(find_dotenv())

# Configurar la página de Streamlit
st.set_page_config(page_icon="📃", layout="wide", page_title="Groq & LLaMA3.1 Chat Bot...")

def icon(emoji: str):
 """Muestra un emoji como ícono de página estilo Notion."""
 st.write(
 f'<span style="font-size: 78px; line-height: 1">{emoji}</span>',
 unsafe_allow_html=True,
 )

# Encabezado de la aplicación
st.subheader("Groq Chat with LLaMA3.1 App", divider="rainbow", anchor=False)

# Inicializar cliente Groq
client = Groq(
 api_key=os.environ['GROQ_API_KEY'],
)

# Inicializar historial de chat y modelo seleccionado
if "messages" not in st.session_state:
 st.session_state.messages = []

if "selected_model" not in st.session_state:
 st.session_state.selected_model = None

# Detalles de los modelos
models = {
 "llama-3.1-70b-versatile": {"name": "LLaMA3.1-70b", "tokens": 4096, "developer": "Meta"},
 "llama-3.1-8b-instant": {"name": "LLaMA3.1-8b", "tokens": 4096, "developer": "Meta"},
 "llama3-70b-8192": {"name": "Meta Llama 3 70B", "tokens": 4096, "developer": "Meta"},
 "llama3-8b-8192": {"name": "Meta Llama 3 8B", "tokens": 4096, "developer": "Meta"},
 "llama3-groq-70b-8192-tool-use-preview": {"name": "Llama 3 Groq 70B Tool Use (Preview)", "tokens": 4096, "developer": "Groq"},
 "gemma-7b-it": {"name": "Gemma-7b-it", "tokens": 4096, "developer": "Google"},
 "mixtral-8x7b-32768": {
 "name": "Mixtral-8x7b-Instruct-v0.1",
 "tokens": 32768,
 "developer": "Mistral",
 },
}

# Diseño para la selección de modelo y slider de tokens
col1, col2 = st.columns([1, 3]) # Ajusta la proporción para hacer la primera columna más pequeña

with col1:
 model_option = st.selectbox(
 "Choose a model:",
 options=list(models.keys()),
 format_func=lambda x: models[x]["name"],
 index=0, # Predeterminado al primer modelo en la lista
 )
 max_tokens_range = models[model_option]["tokens"]
 max_tokens = st.slider(
 "Max Tokens:",
 min_value=512,
 max_value=max_tokens_range,
 value=min(32768, max_tokens_range),
 step=512,
 help=f"Adjust the maximum number of tokens (words) for the model's response. Max for selected model: {max_tokens_range}",
 )

# Detectar cambio de modelo y limpiar historial de chat si el modelo ha cambiado
if st.session_state.selected_model != model_option:
 st.session_state.messages = []
 st.session_state.selected_model = model_option

# Añadir un botón para "Limpiar Chat"
if st.button("Clear Chat"):
 st.session_state.messages = []

# Mostrar mensajes de chat del historial en la aplicación
for message in st.session_state.messages:
 avatar = "🔋" if message["role"] == "assistant" else "🧑‍💻"
 with st.chat_message(message["role"], avatar=avatar):
 st.markdown(message["content"])

def generate_chat_responses(chat_completion) -> Generator[str, None, None]:
 """Generar contenido de respuesta del chat a partir de la respuesta de la API de Groq."""
 for chunk in chat_completion:
 if chunk.choices[0].delta.content:
 yield chunk.choices[0].delta.content

# Instrucción privada que se aplicará a cada mensaje
private_instruction = (
 "# Extract the benefits of the product, not the features. # You should be as brief as possible. # Omit the price, if any. # Do not mention the name of the product. # Use 3 paragraphs. # Try to synthesize or summarize. # Focus only on the benefits. # Highlight how this product helps the customer. # Always respond in Spanish. # The text you create will be used in an e-commerce product sales page through the Internet, so it must be persuasive, attractive, and above all very short and summarized. # Remember to keep the text short, summarized, synthesized in three paragraphs. # Surprise me with your best ideas! # Always answers in AMERICAN SPANISH. Stop after finish the first content genreated."
)

# Manejar la entrada del chat del usuario
if prompt := st.chat_input("Escribe tu mensaje aquí..."):
 st.session_state.messages.append({"role": "user", "content": prompt})

 with st.chat_message("user", avatar="🧑‍💻"): 
 st.markdown(prompt)

 # Preparar los mensajes para la API, incluyendo la instrucción privada
 messages_for_api = [
 {"role": "system", "content": private_instruction},
 ] + [
 {"role": m["role"], "content": m["content"]}
 for m in st.session_state.messages
 ]

 # Obtener respuesta de la API de Groq
 try:
 chat_completion = client.chat.completions.create(
 model=model_option,
 messages=messages_for_api,
 max_tokens=max_tokens,
 stream=True,
 )

 # Usar la función generadora con st.write_stream
 with st.chat_message("assistant", avatar="🔋"):
 chat_responses_generator = generate_chat_responses(chat_completion)
 full_response = st.write_stream(chat_responses_generator)

 # Añadir la respuesta completa al historial de mensajes
 if isinstance(full_response, str):
 st.session_state.messages.append(
 {"role": "assistant", "content": full_response}
 )
 else:
 combined_response = "\n".join(str(item) for item in full_response)
 st.session_state.messages.append(
 {"role": "assistant", "content": combined_response}
 )

 except Exception as e:
 st.error(e, icon="❌")