import streamlit as st
from transformers import pipeline
import torch

# Page Config
st.set_page_config(page_title='Transformers in NLP', layout='wide')

# Page Title
st.markdown('<h1 style="color:#4CAF50; text-align:center;">🤖 Transformers & Pretrained Models in NLP 🚀</h1>', unsafe_allow_html=True)

# Transformer Architecture
st.markdown('<h2 style="color:#FF5733">📌 1. Transformer Architecture</h2>', unsafe_allow_html=True)

st.subheader('🔎 Definition:')
st.write("""
The **Transformer architecture** revolutionized NLP by using **self-attention** to process sequences in parallel. 
- **Self-attention** enables words to focus on others dynamically.
- The **encoder-decoder** structure is used in tasks like translation.

📜 Introduced in "**Attention is All You Need**" (Vaswani et al., 2017).
""")

st.subheader('🛠️ Key Components:')
st.write("""
- **Encoder**: Processes input tokens into internal representations.
- **Decoder**: Uses encoder outputs to generate predictions.
- **Multi-head Attention**: Allows diverse focus across sequences.
- **Positional Encoding**: Injects sequence order into embeddings.
""")

# Pretrained Models
st.markdown('<h2 style="color:#3E7FCB">📌 2. Pretrained Models</h2>', unsafe_allow_html=True)

st.subheader('🔍 Definition:')
st.write("""
Pretrained models leverage vast corpora to understand language patterns.
- **BERT**: Bi-directional context learning for diverse NLP tasks.
- **GPT**: Text generation with autoregressive modeling.
- **RoBERTa**: Optimized BERT variant.
- **T5**: Universal text-to-text learning.
- **XLNet**: Captures dependencies in all positions.
""")

# Sentiment Analysis Example
st.subheader('🎭 Pretrained Model Example: Sentiment Analysis')
nlp = pipeline("sentiment-analysis", model="bert-base-uncased")
text = st.text_area("📝 Enter text to analyze", "Transformers are amazing!")
if st.button('📊 Analyze Sentiment'):
    result = nlp(text)
    st.write(f"**🧐 Result:** {result}")

# Fine-tuning Pretrained Models
st.markdown('<h2 style="color:#E67E22">📌 3. Fine-tuning Pretrained Models</h2>', unsafe_allow_html=True)

st.subheader('⚙️ Definition:')
st.write("""
Fine-tuning tailors pretrained models to specific NLP tasks:
- **Sentiment Analysis**: Classifies text sentiments.
- **Named Entity Recognition (NER)**: Detects names, locations, organizations.
- **Question Answering**: Extracts answers from given contexts.
""")

# NER Example
st.subheader('🔎 Named Entity Recognition (NER)')
nlp_ner = pipeline("ner", model="dbmdz/bert-large-cased-finetuned-conll03-english")
text_ner = st.text_area("🔤 Enter text for NER", "Barack Obama was born in Hawaii.")
if st.button('🔬 Perform NER'):
    ner_results = nlp_ner(text_ner)
    st.write("**🔍 NER Results:**")
    for entity in ner_results:
        st.write(f"📝 {entity['word']} - {entity['entity']} - Confidence: {entity['score']:.2f}")

# Question Answering Example
st.subheader('🧠 Question Answering with BERT')
nlp_qa = pipeline("question-answering", model="bert-large-uncased-whole-word-masking-finetuned-squad")
context = st.text_area("📚 Enter context", "Transformers revolutionized NLP with parallel processing.")
question = st.text_input("❓ Ask a question", "What did transformers revolutionize?")
if st.button('🎤 Get Answer'):
    answer = nlp_qa(question=question, context=context)
    st.write(f"**📝 Answer:** {answer['answer']}")

st.markdown('<h3 style="color:#4CAF50; text-align:center;">✨ Thanks for Exploring NLP! ✨</h3>', unsafe_allow_html=True)