abstractive_summarization.py

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline

# Load the fine-tuned BART tokenizer and model
tokenizer = AutoTokenizer.from_pretrained("EE21/BART-ToSSimplify")
model = AutoModelForSeq2SeqLM.from_pretrained("EE21/BART-ToSSimplify")

# Load the BART-large-cnn tokenizer and model
tokenizer_cnn = AutoTokenizer.from_pretrained("facebook/bart-large-cnn")
model_cnn = AutoModelForSeq2SeqLM.from_pretrained("facebook/bart-large-cnn")

# Function to summarize using the fine-tuned BART model
def summarize_with_bart_ft(input_text):
    inputs = tokenizer.encode("summarize: " + input_text, return_tensors="pt", max_length=1024, truncation=True)
    summary_ids = model.generate(inputs, max_length=300, min_length=100, num_beams=1, early_stopping=False, length_penalty=1)
    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return summary

# Function to summarize using bart-large-cnn model
def summarize_with_bart_cnn(input_text):
    inputs = tokenizer_cnn.encode("summarize: " + input_text, return_tensors="pt", max_length=1024, truncation=True)
    summary_ids = model_cnn.generate(inputs, max_length=300, min_length=100, num_beams=1, early_stopping=True, length_penalty=1)
    summary = tokenizer.decode(summary_ids[0], skip_special_tokens=True)
    return summary

# Function to summarize using led-base-book-summary
def summarize_with_led(input_text):
    pipe_led = pipeline("summarization", model="pszemraj/led-base-book-summary")
    summary = pipe_led(input_text, max_length=300, min_length=100, num_beams=1, early_stopping=False, length_penalty=1)
    return summary[0]['summary_text']

# Function to summarize using long-t5-tglobal-base-sci-simplify
def summarize_with_t5(input_text):
    pipe_t5 = pipeline("summarization", model="pszemraj/long-t5-tglobal-base-sci-simplify")
    summary = pipe_t5(input_text, max_length=300, min_length=100, num_beams=1, early_stopping=False, length_penalty=1)
    return summary[0]['summary_text']