Totto_t5_base_BERT_Score_20k_steps

language:

• en

tags:

• Table to text
• Data to text

Dataset:

• ToTTo A Controlled Table-to-Text Dataset. Totto is an open-source table-to-text dataset with over 1,20,000 examples in the English language. It defines a controlled generation task as: given a Wikipedia table and a set of highlighted cells, generate a one-sentence description.

Base Model - T5-Base

Google's T5 The T5 was built by the Google team in order to create a general-purpose model that can understand the text. The basic idea behind t5 was to deal with the text processing problem as a “text-to-text” problem, i.e. taking the text as input and producing new text as output.

Baseline Preprocessing

Baseline Preprocessing This code repository serves as a supplementary for the main repository, which can be used to do basic preprocessing of the Totto dataset.

Fine-tuning

On the Totto dataset, we used the T5 for the conditional generation model and fine-tuned it with 10000 steps BLEU and then 20000 steps BERT-SCORE as a metric.