Text-Summarizer / app.py
Gladiator's picture
cache load model funcs for faster load times
4065f3f
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 πŸ“")
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)