nassimjp/pashto-restaurant-chatbot

🇦🇫 Pashto Restaurant Chatbot Training Dataset (With Love for Pashto AI)

بیا رغونه او پښتو ژبې ته ځانګړې پاملرنه؛ د پښتو مصنوعي ځیرکتیا (Pashto AI) د بډاینې او ودې لپاره په مینه چمتو شوی کڅوړه.

This dataset is a high-quality, state-resilient Pashto translation of the widely used bitext/Bitext-restaurants-llm-chatbot-training-dataset. It contains approximately 30,000 conversational instruction-response pairs meticulously optimized for domain-specific fine-tuning in the hospitality, restaurant, and customer support verticals.

---

📊 Dataset Properties

• Source Dataset: Bitext Restaurants LLM Chatbot Training Dataset
• Primary Target Language: Pashto (پښتو)
• Format: JSON Lines (.jsonl)
• Key Feature: 100% strict preservation of CRM template placeholders such as {{WEBSITE_URL}}, {{ORDERS_SECTION}}, and {{CANCEL_ORDER_OPTION}}.

---

📂 Data Sample Structure

Each entry features the native user intent configuration mapped alongside the synchronized structured multi-line support response:

{
  "instruction": "زما پلانونه بدل شوي، زه څنګه کولی شم خپل د کیټرینګ آرډر لغوه کړم؟",
  "response": "موږ پوهیږو چې شرایط ممکن په ناڅاپي ډول بدل شي، او موږ چمتو یو چې ستاسو د خواړو امر لغوه کولو کې ستاسو سره مرسته وکړو.\n\n1. خپل حساب ته په {{WEBSITE_URL}} کې لاسرسی ومومئ.\n2. {{ORDERS_SECTION}} ساحې ته لاړ شئ.\n3. د خوړو ځانګړي امر ومومئ چې تاسو یې لغوه کول غواړئ او په هغې کلیک وکړئ.\n4. د {{CANCEL_ORDER_OPTION}} انتخاب غوره کړئ او د لغوه کولو پریکړه تایید کړئ.\n\nکه تاسو نور ملاتړ ته اړتیا لرئ، د مرستې لپاره زموږ د پیرودونکي خدماتو ټیم سره اړیکه ونیسئ.",
  "_original_hash": "b2c7e8913d8d646b9a89c923bb6f18d2"
}

---

🛠️ Pipeline & Engineering Integrity

To ensure maximum performance when training tokenizer configurations and local LLMs, our translation framework integrated the following constraints:

1. Multi-Byte Script Verification: String validation routines explicitly verified the presence of native Pashto-specific alphabets (ړ, ې, ښ, څ, ۍ, ټ, ډ, ږ, ځ).
2. Contextual Token Segmentation: Sentences were parsed safely below 3,000 characters to prevent semantic fragmentation.
3. No Hidden Artifacts: Completely stripped out corrupted or incorrect fallback translation remnants, providing a pristine dataset card setup.

---

⚖️ License & Attribution (CDLA-Sharing-1.0)

In strict accordance with the original dataset author (Bitext), this derivative work is published under the Community Data License Agreement – Sharing – Version 1.0 (CDLA-Sharing-1.0).

Key Copyleft Obligations:

• Attribution: You must attribute this derivative version to the upstream publisher (bitext) and the localization system.
• ShareAlike: If you compute, modify, scale, or build upon this dataset to release a new public data asset, you must make that resulting data asset available under the same CDLA-Sharing-1.0 terms.
• Computational Use: Results obtained from computational analysis or training models (weights, parameters, checkpoint outputs) do not inherit the data copyleft constraint and are unrestricted.

---

🚀 Quick Start with 🤗 Datasets

from datasets import load_dataset

dataset = load_dataset("nassimjp/pashto-restaurant-chatbot")
print(dataset['train'][0])

Made with 💖 for the Pashto Open Source NLP Community. Let's build robust local AI configurations!