Napolab: Prepare Portuguese LLM benchmark data

Understand the project

Practical overview

Use Napolab when you need Portuguese-language benchmark data or want to review model performance on the associated leaderboard.

Researchers evaluating large language models on Portuguese-language tasks.
Developers preparing Portuguese benchmark datasets for further model experiments.

Browse model results on Portuguese benchmarks.
Prepare benchmark data locally as a CSV in completions format.
Select curated Portuguese datasets for tasks such as entailment, similarity, hate-speech classification and sentence simplification.

Before you choose

The dataset collection and leaderboard are linked through Hugging Face; the source repository is public and licensed under MIT.

The LLM-oriented format pairs each prompt with an expected classification label between 0 and 5.
Dataset inclusion guidelines prioritize natural or professionally translated text, public datasets and expert human annotations.
HateBR is restricted to academic and research purposes unless prior written consent is obtained from SINCH.
The documented local workflow exports and converts benchmark data; it does not describe running models or generating evaluation scores.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Prepare the benchmark locally as a CSV in completions format.

Clone the Napolab repository and change into its directory.
Install the library with the command below.
Documented command
```
pip install -e .
```
Call export_napolab_benchmark() and pass its result to convert_to_completions_format().
Reset the output index and write the dataframe to test.csv.

A local `test.csv` file containing the converted benchmark data.

This workflow prepares benchmark data; the documented steps do not run model evaluations.
The README notes that HateBR is for academic and research use; other use requires prior written consent from SINCH.

Evidence and transparency

Source record details

Source availability: Active
Sources checked: 15 October 2026
Original website title: ruanchaves/napolab · Official homepage title · retained in its source language