T5 Qa Webnlg Synth En

🚀 t5-qa_webnlg_synth-en

A Data Question Answering model based on T5-small that answers questions with a structured table as input.

🚀 Quick Start

This model is a Data Question Answering model based on T5-small, which can answer questions when given a structured table as input. It's actually a part of the QuestEval metric, but it can also be used independently for QA tasks.

✨ Features

• Based on the T5-small architecture.
• Capable of answering questions with a structured table as input.
• Can be used as a standalone component for QA.

📦 Installation

No specific installation steps are provided in the original README. However, here is the code to load the model:

from transformers import T5Tokenizer, T5ForConditionalGeneration

tokenizer = T5Tokenizer.from_pretrained("ThomasNLG/t5-qa_webnlg_synth-en")

model = T5ForConditionalGeneration.from_pretrained("ThomasNLG/t5-qa_webnlg_synth-en")

💻 Usage Examples

Basic Usage

from transformers import T5Tokenizer, T5ForConditionalGeneration

tokenizer = T5Tokenizer.from_pretrained("ThomasNLG/t5-qa_webnlg_synth-en")

model = T5ForConditionalGeneration.from_pretrained("ThomasNLG/t5-qa_webnlg_synth-en")

Advanced Usage

You can play with the model using the inference API. The text input format should follow this template (accordingly to the training stage of the model):

text_input = "{QUESTION} </s> {CONTEXT}"

where CONTEXT is a structured table that is linearised this way:

CONTEXT = "name [ The Eagle ] , eatType [ coffee shop ] , food [ French ] , priceRange [ £ 2 0 - 2 5 ]"

# Example of creating text_input
question = "What is the food type at The Eagle?"
context = "name [ The Eagle ] , eatType [ coffee shop ] , food [ French ] , priceRange [ £ 2 0 - 2 5 ]"
text_input = f"{question} </s> {context}"

input_ids = tokenizer(text_input, return_tensors="pt").input_ids
outputs = model.generate(input_ids)
answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(answer)

📚 Documentation

Training data

The model was trained on synthetic data as described in Data-QuestEval: A Referenceless Metric for Data to Text Semantic Evaluation.

Citation info

@article{rebuffel2021data,
  title={Data-QuestEval: A Referenceless Metric for Data to Text Semantic Evaluation},
  author={Rebuffel, Cl{\'e}ment and Scialom, Thomas and Soulier, Laure and Piwowarski, Benjamin and Lamprier, Sylvain and Staiano, Jacopo and Scoutheeten, Geoffrey and Gallinari, Patrick},
  journal={arXiv preprint arXiv:2104.07555},
  year={2021}
}

📄 License

This project is licensed under the MIT license.