Lilt Document QA

🚀 LiLT Document Question Answering Model

This project focuses on the LiLT model, which is fine - tuned for English DocVQA. It provides a solution for document question - answering tasks.

🚀 Quick Start

The LiLT model is designed for document question - answering. You can find more details about the model here.

💻 Usage Examples

Basic Usage

from transformers import AutoTokenizer, AutoModelForQuestionAnswering
from datasets import load_dataset

model_checkpoint = "TusharGoel/LiLT-Document-QA"
tokenizer = AutoTokenizer.from_pretrained(model_checkpoint, add_prefix_space=True)
model_predict = AutoModelForQuestionAnswering.from_pretrained(model_checkpoint)

model_predict.eval()
dataset = load_dataset("nielsr/funsd", split="train")
example = dataset[0]
print(example)

question = "What is the Licensee Number?"
print(question)

words = example["words"]
boxes = example["bboxes"]

encoding = tokenizer(question, words, boxes = boxes, return_token_type_ids=True, return_tensors="pt")

word_ids = encoding.word_ids(0)
outputs = model_predict(**encoding)

loss = outputs.loss
start_scores = outputs.start_logits
end_scores = outputs.end_logits

start, end = word_ids[start_scores.argmax(-1).item()], word_ids[end_scores.argmax(-1).item()]
# print(start, end)
print(" ".join(words[start : end + 1]))

Advanced Usage

There is no advanced usage example provided in the original document.

📄 License

The project is licensed under the MIT license.