Upload README.md with huggingface_hub

README.md

@@ -18,27 +18,23 @@ tags:
 - pt-PT
 - european-portuguese
 - instruction-tuning
-- conversational-ai
-- chatbot
-- helpdesk
-- llm
-- sft
-- alpaca
 - fine-tuning
 - synthetic
 ---
 
 # LusoSupport-PT Lite
 
-🇵🇹 **The only purpose-built European Portuguese (pt-PT) customer support instruction dataset** — verified against the Hugging Face Hub, no direct competitor found.
+> **Project evolution:** LusoSupport-PT launched as a standalone pt-PT instruction dataset. Since AMALIA (Portugal's open-source, government-backed pt-PT LLM) launched in July 2026, the project has repositioned around specializing AMALIA for customer support — this dataset is preserved and still available, now as the training material behind that specialization. See [the pivot strategy doc](https://github.com/ariazevedopt/pt-ai-instruction-dataset/blob/main/docs/superpowers/specs/2026-08-28-amalia-pivot-strategy.md) for the full rationale.
 
-**LusoSupport-PT** is a structured **European Portuguese (pt-PT)** customer support instruction dataset for LLM fine-tuning, prompting experiments, and support automation.
+🇵🇹 **AMALIA already speaks fluent European Portuguese — LusoSupport-PT specializes it for customer support.** AMALIA (amaliallm.pt) is Portugal's open-source, government-and-university-backed LLM built for European Portuguese; its own instruction data is general-purpose, not support-domain. LusoSupport-PT fills that gap.
+
+**LusoSupport-PT** is a structured **European Portuguese (pt-PT)** customer support instruction dataset for LLM fine-tuning (AMALIA or any other model), prompting experiments, and support automation.
 
 This **Lite** version contains **200 curated rows** — a representative free sample covering all 8 domains, 8 task types, and all 18 customer intents.
 
-The full dataset (10 828 rows) is available for purchase on [Gumroad](https://ariazevedo.gumroad.com/l/lusosupport-pt) — see pricing below.
+The full dataset (10,828 rows) is available for purchase on [Gumroad](https://ariazevedo.gumroad.com/l/lusosupport-pt) — see pricing below.
 
-🔥 **Early Adopter Pricing:** the first 50 buyers on each tier lock in the launch price. After that, prices rise to €79 / €329.
+🔥 **Early Adopter Pricing:** the first 50 buyers on each tier lock in the launch price. After that, prices rise to €39 / €165.
 
 ---
 
@@ -121,6 +117,22 @@ alpaca = [
 ]
 ```
 
+### Fine-tuning AMALIA (ChatML format)
+
+This repo also ships `lusosupport_pt_lite_amalia_chat.jsonl` — the same 200 rows pre-converted to the `messages`/role-content format AMALIA (and other ChatML-style chat models) expect:
+
+```python
+import json
+
+with open("lusosupport_pt_lite_amalia_chat.jsonl") as f:
+    rows = [json.loads(l) for l in f]
+
+print(rows[0]["messages"])
+# [{"role": "system", ...}, {"role": "user", ...}, {"role": "assistant", ...}]
+```
+
+Feed each row's `messages` list into `tokenizer.apply_chat_template()` at training time — see the full walkthrough in [`docs/integration.md` §2](https://github.com/ariazevedopt/pt-ai-instruction-dataset/blob/main/docs/integration.md#2-fine-tuning-amalia-for-customer-support) in the GitHub repo.
+
 ### Intent classification example
 
 ```python
@@ -142,8 +154,8 @@ print(result)
 | Tier | Price | Rows | Licence |
 |---|---|---|---|
 | **Lite (this dataset)** | Free | 200 | CC BY 4.0 |
-| **Premium Individual** | €59 → €79 | 10 828 | Personal / research / non-commercial |
-| **Commercial Licence** | €229 → €329 | 10 828 | Commercial use in products & APIs |
+| **Premium Individual** | €29 → €39 | 5 162 | Personal / research / non-commercial |
+| **Commercial Licence** | €115 → €165 | 5 162 | Commercial use in products & APIs |
 
 👉 [Buy Individual Licence](https://ariazevedo.gumroad.com/l/lusosupport-pt) · [Buy Commercial Licence](https://ariazevedo.gumroad.com/l/lusosupport-pt-commercial) · [GitHub Sponsors](https://github.com/sponsors/ariazevedopt)