README.md

metadata

license: mit
task_categories:
  - text-classification
  - question-answering
  - text-generation
language:
  - en
size_categories:
  - 10M<n<100M

📚 TinyWay-Gutenberg-Clean (Compressed Shards)

A large-scale, high-quality English text dataset derived from Project Gutenberg. The corpus has been cleaned, normalized, deduplicated, segmented into fixed-length samples, and stored as compressed JSONL shards for efficient large-scale language model training.

This dataset is intended for pretraining and experimentation with small and medium language models such as TinyWay, tokenizer training, and large-scale NLP research.

---

📦 Dataset Overview

• Name: TinyWay-Gutenberg-Clean
• Current Release: ~19 compressed shards (.jsonl.gz)
• Estimated Samples: Tens of millions of text segments
• Language: English
• Format: Gzip-compressed JSON Lines (.jsonl.gz)
• Source: Project Gutenberg (public domain books)
• License: Public Domain
• Maintainer: Shivam (NNEngine / ITM AIR Lab)

Each record contains a clean text segment between 30 and 60 words.

Future releases will scale this dataset further (e.g., 100M+ samples).

---

Data Format

Each line is a JSON object:

{
  "id": "twg_000000012345",
  "text": "Cleaned natural English text segment between thirty and sixty words.",
  "word_count": 42,
  "source": "gutenberg"
}

Fields

Field	Description
id	Unique sample identifier
text	Clean English text segment
word_count	Number of words in the segment
source	Data source identifier

---

Data Processing Pipeline

The dataset was generated using a fully streaming pipeline to ensure scalability and low memory usage.

Processing Steps

1. Streaming Input

   • Text streamed from a Project Gutenberg mirror on Hugging Face.

2. Text Cleaning

   • Removed Gutenberg headers and footers.
   • Removed chapter titles, page numbers, and boilerplate text.
   • Normalized whitespace and line breaks.
   • Removed non-ASCII and control characters.
   • Filtered malformed or extremely short segments.

3. Segmentation

   • Text segmented into chunks of 30–60 words.

4. Validation

   • Enforced word count limits.
   • Filtered invalid or noisy segments.

5. Deduplication

   • Exact hash-based deduplication applied during generation.

6. Compression & Sharding

   • Data stored as .jsonl.gz shards for efficient disk usage and streaming.

---

How to Load the Dataset

Using Hugging Face Datasets (Streaming)

from datasets import load_dataset

dataset = load_dataset(
    "NNEngine/TinyWay-Gutenberg-Clean",
    split="train",
    streaming=True
)

for i, sample in enumerate(dataset):
    print(sample)
    if i == 3:
        break

---

Reading a Shard Manually

import gzip
import json

with gzip.open("train-00000.jsonl.gz", "rt", encoding="utf-8") as f:
    for _ in range(3):
        print(json.loads(next(f)))

---

Dataset Characteristics (Approximate)

• Average words per sample: ~45
• Style: Literary and narrative English
• Domain: Fiction, non-fiction, historical texts
• Vocabulary: Large natural English vocabulary
• Compression: ~60–70% size reduction vs raw JSONL

Exact statistics may vary per shard and will be expanded in future releases.

---

Limitations

• Primarily literary and historical language.
• No conversational chat data.
• No code or structured technical documentation.
• Some archaic vocabulary and sentence structures may appear.
• Deduplication is hash-based (near-duplicates may remain).

For conversational or web-style language modeling, this dataset should be mixed with complementary corpora.

---

License

All source texts originate from Project Gutenberg and are in the public domain. This processed dataset is released for unrestricted research and commercial use.

---

Versioning & Roadmap

Planned future updates:

• Larger releases (target: 100M+ samples)
• Improved deduplication (near-duplicate filtering)
• Dataset statistics and analytics
• Additional language normalization

Each major release will be versioned clearly.

---

Citation

If you use this dataset in research or publications, please cite:

TinyWay-Gutenberg-Clean
Shivam (NNEngine), 2026