rejauldu/bn-number-v1

Bengali Numeric Dataset (bn-name-v1)

Dataset Summary

The Bengali Numeric Dataset provides a collection of Bengali number words and arithmetic question–answer pairs.
It is designed for language modeling, tokenization experiments, and fine-tuning models on Bengali numerals.
This dataset includes mappings of integers (0–100, 1000, 1 lakh, 1 crore) into their Bengali text representation and simple arithmetic expressions.

This version (v1) contains train and validation splits (90% train, 10% validation).

---

Supported Tasks and Leaderboards

• Text-to-Text Training: Learn Bengali number representations.
• Arithmetic QA: Train models to answer addition/subtraction questions in Bengali.
• Tokenizer Training: Improve tokenization coverage for Bengali digits and math-related terms.

---

Languages

• Bengali (Bangla)

---

Dataset Structure

Data Fields

• system: Context or system prompt (e.g., date/time, instructions).
• question: User query (e.g., "দুই এর সাথে তিন যোগ করলে কত হবে?").
• answer: Assistant’s reply (e.g., "পাঁচ।").

Splits

• Train: 90% of total dataset
• Validation: 10% of total dataset

---

Dataset Size & Statistics

• Train split: 2,710,143 samples
• Validation split: 301,127 samples
• Total: 3,011,270 samples

Example

{
  "system": "আজকের তারিখ: ২০/০৯/২০২৫\nএখন সময়: সকাল ৯:০০\nআজকের বার: শনিবার",
  "question": "তিন এর সাথে সাত যোগ করলে কত হবে?",
  "answer": "দশ।",
}