File size: 1,983 Bytes
4bcc3a0 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
import torch
import gradio as gr
from transformers import T5Tokenizer, T5ForConditionalGeneration
# Set device
device = "mps" if torch.backends.mps.is_available() else "cuda" if torch.cuda.is_available() else "cpu"
# Load model and tokenizer
model_path = "thenHung/question_decomposer_t5"
tokenizer = T5Tokenizer.from_pretrained(model_path)
model = T5ForConditionalGeneration.from_pretrained(model_path)
model.to(device)
model.eval()
def decompose_question(question):
"""
Decompose a complex question into sub-questions
Args:
question (str): Input complex question
Returns:
list: List of decomposed sub-questions
"""
try:
# Prepare input
input_text = f"decompose question: {question}"
input_ids = tokenizer(
input_text,
max_length=128,
padding="max_length",
truncation=True,
return_tensors="pt"
).input_ids.to(device)
# Generate sub-questions
with torch.no_grad():
outputs = model.generate(
input_ids,
max_length=128,
num_beams=4,
early_stopping=True
)
# Decode and split output
decoded_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
sub_questions = decoded_output.split(" [SEP] ")
return sub_questions
except Exception as e:
return [f"Error: {str(e)}"]
# Create Gradio interface
demo = gr.Interface(
fn=decompose_question,
inputs=gr.Textbox(label="Enter your complex question"),
outputs=gr.JSON(label="Decomposed Sub-Questions"),
title="Question Decomposer",
description="Breaks down complex questions into simpler sub-questions using a T5 model",
examples=[
"Who is taller between John and Mary?",
"What is the capital of Vietnam and the largest city in Vietnam?",
]
)
# Launch the app
if __name__ == "__main__":
demo.launch() |