import gradio as gr
import re
from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Model choices ordered by accuracy
model_choices = {
    "Pegasus (google/pegasus-xsum)": "google/pegasus-xsum",
    "BigBird-Pegasus (google/bigbird-pegasus-large-arxiv)": "google/bigbird-pegasus-large-arxiv",
    "LongT5 Large (google/long-t5-tglobal-large)": "google/long-t5-tglobal-large",
    "BART Large CNN (facebook/bart-large-cnn)": "facebook/bart-large-cnn",
    "ProphetNet (microsoft/prophetnet-large-uncased-cnndm)": "microsoft/prophetnet-large-uncased-cnndm",
    "LED (allenai/led-base-16384)": "allenai/led-base-16384",
    "T5 Large (t5-large)": "t5-large",
    "Flan-T5 Large (google/flan-t5-large)": "google/flan-t5-large",
    "DistilBART CNN (sshleifer/distilbart-cnn-12-6)": "sshleifer/distilbart-cnn-12-6",
    "DistilBART XSum (mrm8488/distilbart-xsum-12-6)": "mrm8488/distilbart-xsum-12-6",
    "T5 Base (t5-base)": "t5-base",
    "Flan-T5 Base (google/flan-t5-base)": "google/flan-t5-base",
    "BART CNN SamSum (philschmid/bart-large-cnn-samsum)": "philschmid/bart-large-cnn-samsum",
    "T5 SamSum (knkarthick/pegasus-samsum)": "knkarthick/pegasus-samsum",
    "LongT5 Base (google/long-t5-tglobal-base)": "google/long-t5-tglobal-base",
    "T5 Small (t5-small)": "t5-small",
    "MBART (facebook/mbart-large-cc25)": "facebook/mbart-large-cc25",
    "MarianMT (Helsinki-NLP/opus-mt-en-ro)": "Helsinki-NLP/opus-mt-en-ro",
    "Falcon Instruct (tiiuae/falcon-7b-instruct)": "tiiuae/falcon-7b-instruct",
    "BART ELI5 (yjernite/bart_eli5)": "yjernite/bart_eli5"
}

model_cache = {}

# Function to clean input text (remove special characters and extra spaces)
def clean_text(input_text):
    # Replace special characters with a space
    cleaned_text = re.sub(r'[^A-Za-z0-9\s]', ' ', input_text)
    # Replace multiple spaces with a single space
    cleaned_text = re.sub(r'\s+', ' ', cleaned_text)
    # Strip leading and trailing spaces
    cleaned_text = cleaned_text.strip()
    return cleaned_text

# Load model and tokenizer
def load_model(model_name):
    if model_name not in model_cache:
        tokenizer = AutoTokenizer.from_pretrained(model_name)
        model = AutoModelForSeq2SeqLM.from_pretrained(model_name)
        model_cache[model_name] = (tokenizer, model)
    return model_cache[model_name]

# Summarize the text using a selected model
def summarize_text(input_text, model_label, char_limit):
    if not input_text.strip():
        return "Please enter some text."

    # Clean the input text by removing special characters and extra spaces
    input_text = clean_text(input_text)

    model_name = model_choices[model_label]
    tokenizer, model = load_model(model_name)

    # Adjust the input format for T5 and FLAN models
    if "t5" in model_name.lower() or "flan" in model_name.lower():
        input_text = "summarize: " + input_text

    inputs = tokenizer(input_text, return_tensors="pt", truncation=True)

    summary_ids = model.generate(
        inputs["input_ids"],
        max_length=20,  # Still approximate; can be tuned per model
        min_length=5,
        do_sample=False
    )

    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return summary[:char_limit]  # Enforce character limit

# Gradio UI
iface = gr.Interface(
    fn=summarize_text,
    inputs=[
        gr.Textbox(lines=6, label="Enter text to summarize"),
        gr.Dropdown(choices=list(model_choices.keys()), label="Choose summarization model", value="Pegasus (google/pegasus-xsum)"),
        gr.Slider(minimum=30, maximum=200, value=65, step=1, label="Max Character Limit")
    ],
    outputs=gr.Textbox(lines=3, label="Summary (truncated to character limit)"),
    title="Multi-Model Text Summarizer",
    description="Summarize text using different Hugging Face models with a user-defined character limit."
)

iface.launch()