Article
pt-PT ≠ pt-BR: why I built a dedicated European Portuguese support dataset
ariazevedo
• • 1Ari Azevedo PRO
ariazevedo
AI & ML interests
None yet
Recent Activity
upvoted an article about 21 hours ago
pt-PT ≠ pt-BR: why I built a dedicated European Portuguese support dataset posted an update 1 day ago
🇵🇹 LusoSupport-PT v1.1 just shipped: 5,162 → 10,828 rows
Most "Portuguese" instruction datasets floating around are actually Brazilian Portuguese (pt-BR). LusoSupport-PT is different: every row is validated European Portuguese (pt-PT) — "palavra-passe" not "senha", "telemóvel" not "celular", "fatura" not "nota fiscal".
What's inside:
→ 10,828 rows across 8 business domains (e-commerce, travel, subscriptions, telecom, SaaS, utilities, billing, marketplace)
→ 8 task types: intent classification, urgency classification, response generation, email replies, summarization, professional rewriting, next-action suggestion, FAQ answering
→ 18 customer intents, 100% pt-PT validated, zero pt-BR vocabulary leakage
→ Ships as JSONL / CSV / Alpaca JSONL / Parquet — drop-in for Unsloth, LLaMA-Factory, Axolotl, or OpenAI fine-tuning
Example row:
```json
{"instruction": "Classifica a intenção principal do cliente e responde apenas com um objeto JSON.", "input": "Mensagem do cliente: \"Bom dia, preciso da fatura referente ao pagamento deste mês.\"", "output": {"intent": "invoice_request", "urgency": "low", "domain": "billing_accounts"}}
```
Free 200-row sample right here on the Hub. Full dataset (individual or commercial licence) on Gumroad — link in the dataset card.
Built for anyone shipping PT-PT support bots, fine-tuning for the Portugal/PALOP market, or benchmarking European Portuguese generation quality. Feedback and PRs welcome 🙌
#pt-PT #NLP #FineTuning #InstructionTuning #CustomerSupport #Portuguese
published an article 1 day ago
pt-PT ≠ pt-BR: why I built a dedicated European Portuguese support datasetOrganizations
None yet