Skip to content

Dataset · nicholasKluge

Instructional corpus in Brazilian Portuguese

nicholasKluge/Pt-Corpus-Instruct

A massive 6.2 billion token text corpus for pre-training and SFT of models in Portuguese.

What's inside

The dataset combines several Portuguese-language datasets from Hugging Face (Brazilian variant of Portuguese) and weighs 80 GB in uncompressed tokenized form. Inside are about 6.2 billion tokens (from 10 to 100 million records), including dialogue and general instruction data. The dataset is designed for initial pre-training of models considering preferences.

What you can do with it

  • Pre-training or continued training of LLMs in Portuguese
  • Performing SFT to create a Portuguese-language chatbot
  • Adapting multilingual models to the Brazilian variant of Portuguese
Tasks
Text generation
Languages
Portuguese
Size
10M–100M rows
Downloads this month
1.4K
Likes
4

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("nicholasKluge/Pt-Corpus-Instruct")