|
|
---
|
|
|
language:
|
|
|
- pt
|
|
|
license: other
|
|
|
license_name: commercial
|
|
|
license_link: LICENSE
|
|
|
tags:
|
|
|
- rlhf
|
|
|
- human-preferences
|
|
|
- dpo
|
|
|
- reward-model
|
|
|
- portuguese
|
|
|
- brazilian
|
|
|
- communication
|
|
|
- tone-alignment
|
|
|
pretty_name: HUBX Human Preference Dataset (PT-BR)
|
|
|
size_categories:
|
|
|
- 100K<n<1M
|
|
|
task_categories:
|
|
|
- text-classification
|
|
|
- text-generation
|
|
|
---
|
|
|
|
|
|
# 🇧🇷 HUBX Human Preference Dataset (PT-BR)
|
|
|
|
|
|
**The largest Portuguese-Brazilian human preference dataset for RLHF/DPO training.**
|
|
|
|
|
|
## 📊 Dataset Statistics
|
|
|
|
|
|
| Metric | Value |
|
|
|
|--------|-------|
|
|
|
| **Total Annotations** | 314,757 |
|
|
|
| **Unique Tasks** | 450 |
|
|
|
| **Human Annotators** | ~600 |
|
|
|
| **Avg. Votes per Task** | ~699 |
|
|
|
| **Language** | Portuguese (Brazil) |
|
|
|
| **Domain** | Communication Quality & Tone |
|
|
|
|
|
|
## 🎯 Why This Dataset?
|
|
|
|
|
|
- **🇧🇷 Native PT-BR**: Collected from Brazilian Portuguese speakers - not translated
|
|
|
- **👥 Real Humans**: ~600 verified annotators with payment records
|
|
|
- **📈 High Agreement**: ~70% inter-annotator agreement (proves human origin)
|
|
|
- **💬 Justifications**: Includes reasoning behind choices
|
|
|
- **🎯 Practical Domain**: Communication quality, empathy, tone, professionalism
|
|
|
|
|
|
## 🌎 Global Community
|
|
|
|
|
|
Our annotator community is expanding rapidly across multiple regions:
|
|
|
|
|
|
- **South America**: Brazil, Argentina, Colombia, Chile, Peru, and more
|
|
|
- **North America**: United States, Mexico
|
|
|
- **Europe**: Portugal, Spain, Germany, UK
|
|
|
- **Asia**: Japan, Philippines
|
|
|
|
|
|
This diversity enables us to create datasets in multiple languages and cultural contexts. **Interested in custom datasets for other languages? Contact us.**
|
|
|
|
|
|
## 📋 Use Cases
|
|
|
|
|
|
- **RLHF** - Train reward models for Portuguese language models
|
|
|
- **DPO** - Direct Preference Optimization for PT-BR
|
|
|
- **Tone Alignment** - Make AI responses sound more natural and empathetic
|
|
|
- **Customer Service AI** - Train chatbots for Brazilian market
|
|
|
- **UX Writing** - Improve AI-generated microcopy
|
|
|
|
|
|
## 📁 Data Format
|
|
|
```json
|
|
|
{
|
|
|
"id": 1,
|
|
|
"question": "Qual frase soa mais educada em um e-mail profissional?",
|
|
|
"option_a": "Poderia, por gentileza, me enviar os dados até amanhã?",
|
|
|
"option_b": "Preciso dos dados até amanhã.",
|
|
|
"category": "professional_tone",
|
|
|
"metadata": {
|
|
|
"votes_a": 587,
|
|
|
"votes_b": 112,
|
|
|
"total_votes": 699,
|
|
|
"preferred": "A",
|
|
|
"agreement_rate": 0.84,
|
|
|
"confidence": "high"
|
|
|
},
|
|
|
"justifications": [
|
|
|
"A palavra 'por gentileza' demonstra mais respeito",
|
|
|
"A opção A é mais cordial e profissional"
|
|
|
]
|
|
|
}
|
|
|
```
|
|
|
|
|
|
## 🔬 Data Collection
|
|
|
|
|
|
- **Platform**: HUBX - Brazilian AI training community
|
|
|
- **Period**: July - December 2024
|
|
|
- **Method**: Paid annotators completing subscription-based tasks
|
|
|
- **Quality Control**: Time-based filtering, justification collection
|
|
|
- **Verification**: Payment records, unique user accounts, varied timestamps
|
|
|
|
|
|
### Why This Proves Human Origin:
|
|
|
|
|
|
1. **Agreement Rate ~70%** - AI would generate 95%+ agreement
|
|
|
2. **Justifications with typos** - Natural Portuguese errors
|
|
|
3. **Payment Records** - Real transactions for annotations
|
|
|
4. **~600 Unique Accounts** - Verified user registrations
|
|
|
5. **Temporal Distribution** - Responses spread over time, not batch-generated
|
|
|
|
|
|
## 💰 Licensing
|
|
|
|
|
|
**This dataset is available for commercial licensing.**
|
|
|
|
|
|
📧 **Contact for pricing**: contatojeffersonmacedo@gmail.com
|
|
|
|
|
|
### Pricing Tiers:
|
|
|
|
|
|
| License Type | Use Case | Price |
|
|
|
|-------------|----------|-------|
|
|
|
| Research | Academic/Non-commercial | Free (with attribution) |
|
|
|
| Startup | < $1M revenue | $4,500 |
|
|
|
| Commercial | Full commercial rights | $9,500 |
|
|
|
| Enterprise | Exclusive rights negotiable | Contact |
|
|
|
|
|
|
## 📬 Contact
|
|
|
|
|
|
- **Email**: contatojeffersonmacedo@gmail.com
|
|
|
- **Twitter/X**: @hubxai
|
|
|
- **Platform**: https://hubx.app
|
|
|
|
|
|
## 📜 Citation
|
|
|
```bibtex
|
|
|
@dataset{hubx_ptbr_preferences_2024,
|
|
|
title={HUBX Human Preference Dataset (PT-BR)},
|
|
|
author={HUBX Community},
|
|
|
year={2024},
|
|
|
publisher={Hugging Face},
|
|
|
language={Portuguese (Brazil)},
|
|
|
size={314757 annotations}
|
|
|
}
|
|
|
```
|
|
|
|
|
|
---
|
|
|
|
|
|
**⚡ Ready to improve your Portuguese AI? Contact us today.** |
