Spaces:
Sleeping
Sleeping
import gradio as gr | |
from transformers import pipeline | |
import random | |
import nltk | |
nltk.download('punkt') | |
# Dictionary that maps the user-friendly model names to their actual names | |
model_names = { | |
"BERT base": "google-bert/bert-base-cased", | |
"DistilBERT base": "distilbert/distilbert-base-cased", | |
"RoBERTa base": "FacebookAI/roberta-base", | |
"BERT finetuned on a dataset for mask filling": "emma7897/bert_one", | |
"DistilBERT finetuned on a dataset for mask filling": "emma7897/distilbert_one", | |
"BERT finetuned on a dataset of stories for children": "emma7897/bert_two", | |
"DistilBERT finetuned on a dataset of stories for children": "emma7897/distilbert_two", | |
} | |
sample_paragraphs = [ | |
"Once upon a time, in a faraway land, there lived a beautiful princess named [MASK]. She was known throughout the kingdom for her [MASK] and immense bravery. One day, while exploring the large forest, she stumbled upon a [MASK] hidden amongst the trees. Curiosity piqued, she ventured inside and discovered a [MASK] filled with treasures beyond imagination. Little did she know, her adventures were just beginning.", | |
"In the city of [MASK], where the streets were always very crowded and the skyscrapers reached for the sky, there was a tall detective named Sam. With a keen eye for detail and a knack for solving mysteries, Sam was the best in the business. When horrific crime shook the city to its core, Sam was called to travel to [MASK]. With determination and a trusty [MASK] by his side, Sam set out to uncover the truth.", | |
"On a remote island in the middle of the [MASK], there stood a blue lighthouse overlooking the turbulent waters. Inside, a keeper tended to the beacon, guiding [MASK] safely to shore. One stormy night, as the waves crashed against the rocks and the wind howled through the [MASK], a ship appeared on the horizon, its sails tattered and its crew in desperate need of help. With nerves of [MASK] and a steady hand, the lighthouse keeper sprang into action, signaling the way to safety.", | |
"In a whimsical village nestled in the [MASK] countryside, there lived an inventor named Zoey. Day and night, Zoey toiled away in her workshop, creating [MASK] that defied imagination. There was no limit to Zoey's creativity. But when a problem threatened to disrupt the peace of the village, Zoey knew it was time to put her [MASK] to the test. With gears whirring and steam hissing, Zoey set out to save the day.", | |
"Meet Emma, a spirited young soul with [MASK] dreams. Emma's eyes sparkle with determination as she envisions herself soaring among the stars as an aspiring [MASK]. She spends her days devouring books about [MASK]. When Emma is not gazing at the stars, you can find her drawing pictures of [MASK].", | |
"Hello! I would like to introduce you to my best friend, [MASK]." | |
] | |
example_models = [ | |
"BERT base", | |
"DistilBERT base", | |
"RoBERTa base", | |
"BERT finetuned on a dataset for mask filling", | |
"DistilBERT finetuned on a dataset for mask filling", | |
"BERT finetuned on a dataset of stories for children", | |
"DistilBERT finetuned on a dataset of stories for children", | |
] | |
# Create a nested list for the examples | |
examples = [[random.choice(example_models), paragraph] for paragraph in sample_paragraphs] | |
def textGenerator(model, userInput): | |
model_name = model_names[model] | |
fill_mask = pipeline("fill-mask", model=model_name) | |
sentences = nltk.sent_tokenize(userInput) | |
processed_sentences = [] | |
if model_name != "FacebookAI/roberta-base": | |
for sentence in sentences: | |
while "[MASK]" in sentence: | |
predictions = fill_mask(sentence, top_k=10) | |
token_strings = [] | |
for prediction in predictions: | |
token_strings.append(prediction['token_str']) | |
selected_token = random.choice(token_strings) | |
sentence = sentence.replace("[MASK]", f"<mark>{selected_token}</mark>", 1) | |
processed_sentences.append(sentence) | |
processedText = " ".join(processed_sentences) | |
if model_name == "FacebookAI/roberta-base": | |
for sentence in sentences: | |
while "[MASK]" in sentence: | |
sentence = sentence.replace("[MASK]", "<mask>", 1) | |
predictions = fill_mask(sentence, top_k=10) | |
token_strings = [] | |
for prediction in predictions: | |
token_strings.append(prediction['token_str']) | |
selected_token = random.choice(token_strings).strip() | |
sentence = sentence.replace("<mask>", f"<mark>{selected_token}</mark>", 1) | |
processed_sentences.append(sentence) | |
processedText = " ".join(processed_sentences) | |
return processedText | |
screen = gr.Interface(fn=textGenerator, inputs=[ | |
gr.Radio(list(model_names.keys()), label="LLM", info="Which LLM would you like to use?"), | |
gr.Textbox(label = "User Input", info="Please enter a paragraph. Replace words that you want the LLM to fill in with [MASK]. Note: there is a limit of one [MASK] per sentence."), | |
], outputs = gr.HTML(label = "Processed Text"), | |
examples = examples, | |
) | |
if __name__ == "__main__": | |
screen.launch() |