app.py

# Para Spaces de Hugging Face, la app principal debe llamarse app.py
# Este archivo es una copia de ejemplo_2_.py

import streamlit as st
from transformers import (
    AutoTokenizer,
    AutoModelForSeq2SeqLM,
    AutoModelForSequenceClassification,
    pipeline
)
import torch

st.set_page_config(page_title="SLM Demo: QA + Sentiment", page_icon="🧠")
st.title("🧠 Small Language Models Demo")
st.markdown("""
Esta app compara tres Small Language Models:
- `flan-t5-small` para responder preguntas.
- `distilBERT` para análisis de sentimiento.
- `distilBERT` fine-tuned para análisis de sentimiento.
""")

# Cargar modelos y tokenizers
@st.cache_resource
def load_models():
    # FLAN-T5 para QA
    flan_name = "google/flan-t5-small"
    flan_tokenizer = AutoTokenizer.from_pretrained(flan_name)
    flan_model = AutoModelForSeq2SeqLM.from_pretrained(flan_name)

    # DistilBERT para clasificación (modelo base)
    distil_name = "distilbert-base-uncased-finetuned-sst-2-english"
    sentiment_analyzer = pipeline("sentiment-analysis", model=distil_name)

    # DistilBERT fine-tuned propio
    custom_name = "juancmamacias/jd-jcms"
    custom_analyzer = pipeline("sentiment-analysis", model=custom_name)

    return flan_tokenizer, flan_model, sentiment_analyzer, custom_analyzer

flan_tokenizer, flan_model, sentiment_analyzer, custom_analyzer = load_models()

# Historial de preguntas
if "history" not in st.session_state:
    st.session_state.history = []

# Entrada del usuario
question = st.text_input("💬 Escribe una pregunta o frase para analizar:")

if st.button("Procesar") and question:
    with st.spinner("Procesando..."):

        # ➤ Respuesta con FLAN-T5
        input_ids = flan_tokenizer(question, return_tensors="pt").input_ids
        outputs = flan_model.generate(input_ids, max_length=50)
        flan_answer = flan_tokenizer.decode(outputs[0], skip_special_tokens=True)

        # ➤ Clasificación con DistilBERT base
        sentiment = sentiment_analyzer(question)[0]
        sentiment_label = sentiment['label']
        sentiment_score = round(sentiment['score'], 3)

        # ➤ Clasificación con DistilBERT fine-tuned propio
        custom_sentiment = custom_analyzer(question)[0]
        custom_label = custom_sentiment['label']
        custom_score = round(custom_sentiment['score'], 3)

        # Guardar en historial
        st.session_state.history.append({
            "question": question,
            "answer": flan_answer,
            "sentiment": f"{sentiment_label} ({sentiment_score})",
            "custom_sentiment": f"{custom_label} ({custom_score})"
        })

# Mostrar historial
if st.session_state.history:
    st.markdown("### 📜 Historial")
    for i, item in enumerate(reversed(st.session_state.history), 1):
        st.markdown(f"""
        **{i}. Entrada:** {item['question']}  
        🧠 **Respuesta (FLAN):** {item['answer']}  
        ❤️ **Sentimiento (base):** {item['sentiment']}  
        💙 **Sentimiento (propio):** {item['custom_sentiment']}  
        ---""")