from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
import torch
import re
import gradio as gr
import os
import docx2txt

tokenizer = AutoTokenizer.from_pretrained("allenai/led-base-16384")
model = AutoModelForSeq2SeqLM.from_pretrained("checkpoint-64840").to("cpu")

def summarize(text_file):
    file_extension = os.path.splitext(text_file.name)[1]
    if file_extension == ".txt":
        # Load text from a txt file
        with open(text_file.name, "r", encoding="utf-8") as f:
            text = f.read()
    elif file_extension == ".docx":
        # Load text from a Word file
        text = docx2txt.process(text_file.name)
    else:
        raise ValueError(f"Unsupported file type: {file_extension}")
    
    input_ids = tokenizer(text, return_tensors="pt").input_ids.to("cpu")
    global_attention_mask = torch.zeros_like(input_ids)
    # set global_attention_mask on first token
    global_attention_mask[:, 0] = 1

    sequences = model.generate(input_ids, global_attention_mask=global_attention_mask).sequences

    summary = tokenizer.batch_decode(sequences)[0]


    
    return text, summary



iface = gr.Interface(
    fn=summarize,
    inputs=gr.inputs.File(label="Upload a txt file or a Word file for the input text"),
    outputs=[gr.outputs.Textbox(label="Original text"), gr.outputs.Textbox(label="Summary")],
    title="Academic Paper Summarization Demo",
    description="Upload a txt file or a Word file for the input text. Get a summary generated by a small T5 model from Hugging Face.",
)

iface.launch()