Datasets:

Yxanul
/

wikipedia-2k-high-quality

Error code:   FeaturesError
Exception:    ArrowInvalid
Message:      Schema at index 1 was different: 
original_articles: int64
original_tokens: int64
pruned_articles: int64
pruned_tokens: int64
removed_articles: int64
removed_tokens: int64
vs
text: string
token_count: int64
wikidata_id: string
version_id: string
extracted_title: string
Traceback:    Traceback (most recent call last):
                File "/src/services/worker/src/worker/job_runners/split/first_rows.py", line 228, in compute_first_rows_from_streaming_response
                  iterable_dataset = iterable_dataset._resolve_features()
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 3422, in _resolve_features
                  features = _infer_features_from_batch(self.with_format(None)._head())
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 2187, in _head
                  return next(iter(self.iter(batch_size=n)))
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 2391, in iter
                  for key, example in iterator:
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 1882, in __iter__
                  for key, pa_table in self._iter_arrow():
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 1904, in _iter_arrow
                  yield from self.ex_iterable._iter_arrow()
                File "/src/services/worker/.venv/lib/python3.9/site-packages/datasets/iterable_dataset.py", line 543, in _iter_arrow
                  yield new_key, pa.Table.from_batches(chunks_buffer)
                File "pyarrow/table.pxi", line 4116, in pyarrow.lib.Table.from_batches
                File "pyarrow/error.pxi", line 154, in pyarrow.lib.pyarrow_internal_check_status
                File "pyarrow/error.pxi", line 91, in pyarrow.lib.check_status
              pyarrow.lib.ArrowInvalid: Schema at index 1 was different: 
              original_articles: int64
              original_tokens: int64
              pruned_articles: int64
              pruned_tokens: int64
              removed_articles: int64
              removed_tokens: int64
              vs
              text: string
              token_count: int64
              wikidata_id: string
              version_id: string
              extracted_title: string

Need help to make the dataset viewer work? Make sure to review how to configure the dataset viewer, and open a discussion for direct support.

Wikipedia High-Quality Articles (2000+ Tokens)

Dataset Description

This dataset contains high-quality Wikipedia articles filtered to include only substantial, comprehensive content. Each article contains at least 2,000 tokens, ensuring rich, interconnected knowledge rather than simple definitions or stubs.

Key Features

Quality over Quantity: 239K comprehensive articles instead of 1.26M mixed-quality articles
Rich Content: Average article length of 4,032 tokens
Interconnected Knowledge: Articles contain multiple concepts with context and relationships
Clean Format: Structured with article/section/paragraph markers for easy parsing

Dataset Statistics

Metric	Value
Total Articles	238,758
Total Tokens	0.96B
Average Tokens/Article	4032
Min Tokens/Article	2,000
Original Dataset Size	1,259,957 articles
Filtering Rate	81.1% removed
Token Retention	49.5%

Why This Filtering?

Research shows that training on fewer, higher-quality examples for more epochs produces better results than training on more, lower-quality examples:

Conceptual Density: Long articles teach multiple interconnected concepts
Context: Facts are presented with relationships and background
Language Quality: Complex syntax, varied vocabulary, sophisticated discourse
Efficiency: 20 epochs on this dataset > 10 epochs on full dataset

What Was Removed

Simple definitions ("An anode is...")
Lists without context ("List of villages in...")
Stubs and incomplete articles
Single-fact biographical entries

What Was Kept

Comprehensive topic overviews
Historical events with context
Scientific concepts with explanations
Technical topics with examples

Dataset Structure

Each line in the JSONL files contains:

{
  "text": "Full article text with _START_ARTICLE_, _START_SECTION_, _START_PARAGRAPH_ markers",
  "title": "Article title",
  "token_count": 4032,
  "url": "Wikipedia URL"
}

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("Yxanul/wikipedia-2k-high-quality")

# Example: Load for training
for article in dataset['train']:
    text = article['text']
    tokens = article['token_count']
    # Your training code here

Training Recommendations

Based on our research, we recommend:

10-20 epochs for base model pretraining
Sequence length: 2048-4096 tokens (articles are long)
Batch size: Can be smaller due to longer sequences
Curriculum: Start with shorter articles (2K) and progress to longer (10K+)

Source and Processing

This dataset is derived from Wiki40b English dataset, filtered for quality:

Original: Wiki40b English (2.93M articles)
First filter: Removed articles <500 tokens
This version: Kept only articles ≥2000 tokens
Result: Top 19% of articles containing 50% of tokens

License

This dataset inherits Wikipedia's CC BY-SA 3.0 license.

Citation

If you use this dataset, please cite:

@dataset{wikipedia_2k_hq_2024,
  title={Wikipedia High-Quality Articles (2000+ Tokens)},
  author={Yxanul},
  year={2025},
  publisher={HuggingFace}
}

Philosophy

"It's better to deeply understand 239K comprehensive articles than to superficially memorize 1.26M fragments."

This dataset embodies the principle that in machine learning, especially for smaller models, quality dramatically outweighs quantity.

Downloads last month: 81