nelsondiasandre/portuguese-qa-instruct-500

Portuguese Q&A Instruction Dataset (500 pairs)

500 Portuguese (PT-PT) question-answer pairs formatted for instruction fine-tuning of language models.

Dataset Structure

Each example has three columns:

Instruction template

The text column uses the Qwen ChatML format:

<|im_start|>user
{instruction}<|im_end|>
<|im_start|>assistant
{response}<|im_end|>

Dataset Creation

Questions and answers were written in PT-PT (European Portuguese) across 20+ categories:

Splits

Split was performed with seed=42.

Usage

from datasets import load_dataset

ds = load_dataset("nelsondiasandre/portuguese-qa-instruct-500")

# Access a training example
print(ds["train"][0])
# {
#   "instruction": "Qual e a capital de Portugal?",
#   "response": "A capital de Portugal e Lisboa.",
#   "text": "<|im_start|>user\nQual e a capital de Portugal?<|im_end|>\n<|im_start|>assistant\nA capital de Portugal e Lisboa.<|im_end|>"
# }

Intended Use

Supervised fine-tuning (SFT) of language models for Portuguese question answering, particularly with instruction-following models like Qwen2.5.

See the model card for the LoRA adapter trained on this dataset.

Limitations

• —500 examples is a small dataset — models trained on it may overfit or lack coverage
• —PT-PT only (European Portuguese) — not validated for Brazilian Portuguese
• —Answers are intentionally short and direct — not suitable for training long-form generation

License

CC-BY-4.0