import torch import streamlit as st from extractive_summarizer.model_processors import Summarizer from transformers import T5Tokenizer, T5ForConditionalGeneration, T5Config def abstractive_summarizer(text : str, model): tokenizer = T5Tokenizer.from_pretrained('t5-large') device = torch.device('cpu') preprocess_text = text.strip().replace("\n", "") t5_prepared_text = "summarize: " + preprocess_text tokenized_text = tokenizer.encode(t5_prepared_text, return_tensors="pt").to(device) # summmarize summary_ids = abs_model.generate(tokenized_text, num_beams=4, no_repeat_ngram_size=2, min_length=30, max_length=100, early_stopping=True) abs_summarized_text = tokenizer.decode(summary_ids[0], skip_special_tokens=True) return abs_summarized_text @st.cache() def load_ext_model(): model = Summarizer() return model @st.cache() def load_abs_model(): model = T5ForConditionalGeneration.from_pretrained('t5-large') return model if __name__ == "__main__": # --------------------------------- # Main Application # --------------------------------- st.title("Text Summarizer 📝") summarize_type = st.sidebar.selectbox("Summarization type", options=["Extractive", "Abstractive"]) inp_text = st.text_input("Enter the text here") # view summarized text (expander) with st.expander("View input text"): st.write(inp_text) summarize = st.button("Summarize") # called on toggle button [summarize] if summarize: if summarize_type == "Extractive": # extractive summarizer ext_model = load_ext_model() summarized_text = ext_model(inp_text, num_sentences=5) elif summarize_type == "Abstractive": abs_model = load_abs_model() summarized_text = abstractive_summarizer(inp_text, model=abs_model) # final summarized output st.subheader("Summarized text") st.info(summarized_text)