alirani
feat: add pre-synopsis
ff377ee
raw
history blame
4.18 kB
import streamlit as st
import tensorflow as tf
from transformers import AutoTokenizer, TFAutoModelForCausalLM, TFAutoModelForSequenceClassification
# MODEL TO CALL
generator_name = "Alirani/distilgpt2-finetuned-synopsis-genres_final"
tokenizer_gen = AutoTokenizer.from_pretrained(generator_name)
model_gen = TFAutoModelForCausalLM.from_pretrained(generator_name)
def generate_synopsis(model, tokenizer, title):
input_ids = tokenizer(title, return_tensors="tf")
output = model.generate(input_ids['input_ids'], max_length=150, num_beams=5, no_repeat_ngram_size=2, top_k=50, attention_mask=input_ids['attention_mask'])
synopsis = tokenizer.decode(output[0], skip_special_tokens=True)
processed_synopsis = "".join(synopsis.split('|')[2].rpartition('.')[:2]).strip()
return processed_synopsis
classifier_name = "Alirani/overview_classifier_final"
tokenizer_clf = AutoTokenizer.from_pretrained(classifier_name)
model_clf = TFAutoModelForSequenceClassification.from_pretrained(classifier_name)
def generate_classification(model, tokenizer, title, overview):
tokens = tokenizer(f"{title} | {overview}", padding=True, truncation=True, return_tensors="tf")
output = model(**tokens).logits
predicted_class_id = int(tf.math.argmax(output, axis=-1)[0])
return model.config.id2label[predicted_class_id]
favicon = "https://i.ibb.co/JRdhFZg/favicon-32x32.png"
st.set_page_config(page_title="Synopsis Generator", page_icon = favicon, layout = 'wide', initial_sidebar_state = 'auto')
st.title('Demo of a Synopsis Classifier & Generator')
functionality = st.radio("Choose the function to use",
["Classification", "Generation"],
captions = ['Classify title & synopsis into genres', 'Generate synopsis from title & genres'],
index = None)
if functionality == "Classification" :
# CLASSIFY A SYNOPSIS
st.header('Classify a story')
prod_title = st.text_input('Type a title to classify a synopsis')
prod_synopsis = st.text_area('Type a synopsis to classify it')
button_classify = st.button('Get genre')
if button_classify:
if (len(prod_title.split(' ')) > 0) & len(prod_synopsis.split(' ')) > 0:
classified_genre = generate_classification(model_clf, tokenizer_clf, prod_title, prod_synopsis)
st.write('The genre of the title & synopsis is : ', classified_genre)
else:
st.write('Write a title & synopsis for the classifier to work !')
elif functionality == "Generation":
# GENERATE A SYNOPSIS
st.header('Generate a story')
prod_title = st.text_input('Type a title to generate a synopsis')
option_genres = st.selectbox(
'Select a genre to tailor your synopsis',
('Family', 'Romance', 'Comedy', 'Action', 'Documentary', 'Adventure', 'Drama', 'Mystery', 'Crime', 'Thriller', 'Science Fiction', 'History', 'Music', 'Western', 'Fantasy', 'TV Movie', 'Horror', 'Animation', 'Reality'),
index=None,
placeholder="Select genres..."
)
complete_synopsis = st.toggle('Synopsis completion')
if complete_synopsis:
pre_synopsis = st.text_area('Type the beginning of your synopsis')
button_synopsis = st.button('Get synopsis')
if button_synopsis:
if (len(prod_title.split(' ')) > 0) & option_genres:
gen_synopsis = generate_synopsis(model_gen, tokenizer_gen, f"{prod_title} | {option_genres} | {pre_synopsis}")
st.text_area('Generated synopsis', value=gen_synopsis, disabled=True)
else:
st.write('Write a title & select a genre for the generator to work !')
else:
button_synopsis = st.button('Get synopsis')
if button_synopsis:
if len(prod_title.split(' ')) > 0:
gen_synopsis = generate_synopsis(model_gen, tokenizer_gen, f"{prod_title} | {option_genres} | ")
st.text_area('Generated synopsis', value=gen_synopsis, disabled=True)
else:
st.write('Write a title for the generator to work !')
else:
st.write("Select a functionality ! 😊")