Datasets:

Minuri
/

sinhala-validation-set-10k

language:
  - si
license: cc-by-sa-4.0
task_categories:
  - text-generation
pretty_name: Sinhala Validation Set (10K)
size_categories:
  - 10K<n<100K
tags:
  - sinhala
  - low-resource
  - evaluation
  - validation
  - domain-classified

Sinhala Validation Set - 10K Sentences

A held-out Sinhala validation set of 10,000 sentences drawn from the Minuri/diverse_sinhala_dataset corpus. Used for monitoring validation loss during continual pretraining of three LLaMA 3.2 1B variants (Models A, B, C) as part of a diversity-driven Sinhala language model adaptation study.

Dataset Description

This validation set was held out strictly from all three pretraining corpora (A, B, C) to enable unbiased validation loss monitoring during training. It covers multiple domains as classified by a fine-tuned XLM-RoBERTa domain classifier (94% macro-F1).

Source Datasets (via parent corpus)

Source	Description
`madlad`	MADLAD-400 multilingual dataset (Sinhala subset)
`culturax`	CulturaX multilingual web corpus (Sinhala subset)
`nsina`	NSina Sinhala news corpus
`wikipedia`	Sinhala Wikipedia dump

Dataset Structure

Column	Type	Description
`orig_index`	int	Original index in the parent corpus
`sentence`	string	Sinhala sentence text
`source`	string	Source dataset identifier
`predicted_domain`	string	Domain label predicted by XLM-RoBERTa classifier
`confidence`	float	Classifier confidence score

Splits

Split	Rows
train	10,000

Format

Available in both JSONL and CSV formats.

Intended Uses

Validation loss monitoring during Sinhala LLM pretraining
Held-out benchmark for pretraining experiments

Related Repositories

Repo	Description
`Minuri/diverse_sinhala_dataset`	Full 12.38M sentence parent corpus
`Minuri/sinhala-test-set-50k`	Test set (50K sentences)
`Minuri/sinhala-corpus-a-news-1m`	Corpus A - news-only (1M)
`Minuri/sinhala-corpus-b-random-1m`	Corpus B - random (1M)
`Minuri/sinhala-corpus-c-diverse-1m`	Corpus C - diversity-optimised (1M)

Sources & Licenses

This dataset contains sentences derived from the following source datasets. Users must comply with the license terms of each:

Source	License
MADLAD-400	ODC-BY
CulturaX	mC4 + OSCAR licenses
wikimedia/wikipedia	CC BY-SA 3.0 + GFDL
sinhala-nlp/NSINA	CC BY-SA 4.0

This dataset is released under CC BY-SA 4.0 in compliance with the ShareAlike terms of Wikipedia and NSINA.