app.py

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

# Set page configuration
st.set_page_config(page_title="Gemma Paraphraser", page_icon="✍️")

# Load model and tokenizer
@st.cache_resource
def load_model():
    model_name = "EmTpro01/gemma-paraphraser-16bit"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(
        model_name, 
        device_map="cpu",
        torch_dtype=torch.float16
    )
    return model, tokenizer

# Paraphrase function
def paraphrase_text(text, model, tokenizer):
    # Prepare the prompt using Alpaca format
    system_prompt = "Below is provided a paragraph, paraphrase it"
    prompt = f"{system_prompt}\n\n### Input:\n{text}\n\n### Output:\n"
    
    # Tokenize input
    inputs = tokenizer(prompt, return_tensors="pt", add_special_tokens=True)
    
    # Generate paraphrased text
    outputs = model.generate(
        inputs.input_ids, 
        max_length=512,  # Adjust based on your needs
        num_return_sequences=1,
        temperature=0.7,
        do_sample=True
    )
    
    # Decode and clean the output
    paraphrased = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Extract the output part (after "### Output:")
    output_start = paraphrased.find("### Output:") + len("### Output:")
    paraphrased_text = paraphrased[output_start:].strip()
    
    return paraphrased_text

# Streamlit App
def main():
    st.title("📝 Gemma Paraphraser")
    st.write("Paraphrase your text using the Gemma model")

    # Load model
    try:
        model, tokenizer = load_model()
    except Exception as e:
        st.error(f"Error loading model: {e}")
        return

    # Input text area
    input_text = st.text_area("Enter text to paraphrase:", height=200)

    # Paraphrase button
    if st.button("Paraphrase"):
        if input_text:
            with st.spinner("Generating paraphrase..."):
                try:
                    paraphrased_text = paraphrase_text(input_text, model, tokenizer)
                    
                    # Display results
                    st.subheader("Paraphrased Text:")
                    st.write(paraphrased_text)
                    
                    # Optional: Copy to clipboard
                    st.button("Copy to Clipboard", 
                              on_click=lambda: st.write(paraphrased_text))
                except Exception as e:
                    st.error(f"Error during paraphrasing: {e}")
        else:
            st.warning("Please enter some text to paraphrase.")

    # Additional information
    st.sidebar.info(
        "Model: EmTpro01/gemma-paraphraser-16bit\n\n"
        "Tips:\n"
        "- Enter a paragraph to paraphrase\n"
        "- Click 'Paraphrase' to generate\n"
        "- Running on CPU with 16-bit precision"
    )

if __name__ == "__main__":
    main()