The Pile

Combines 22 discrete, curated text datasets (academic papers, books, code, web text, specialized sources) into a single 825 GiB jsonlines corpus compressed with zstandard. The assembly approach prioritizes diversity across domains rather than size maximization, enabling language models trained on this corpus to develop broad cross-domain knowledge and generalization capabilities. Data is provided as-is without documented preprocessing, deduplication, or filtering pipelines, placing responsibility for data cleaning on downstream users.

Provides a standardized evaluation metric (Pile Bits Per Byte, or BPB) that measures language model perplexity across the full 22-subset corpus, enabling comparison of model generalization across diverse text domains. The metric is computed by evaluating a trained model on held-out portions of each subset and aggregating results, producing a single scalar score where lower values indicate better cross-domain performance. This approach surfaces domain-specific weaknesses that single-domain metrics would miss.

Provides training data in a model-agnostic jsonlines format that integrates with standard ML frameworks (PyTorch, TensorFlow, Hugging Face) without requiring custom preprocessing or format conversion. The jsonlines + zstandard approach enables seamless integration with existing dataloaders, tokenizers, and training pipelines, reducing friction for researchers adopting the dataset. No custom APIs or proprietary tools are required — standard open-source libraries suffice.

Encodes the 825 GiB corpus as jsonlines (one JSON object per line, typically with a 'text' field containing raw text) and compresses with zstandard (zstd), a modern compression algorithm offering faster decompression and better compression ratios than gzip. This format choice enables streaming decompression and line-by-line parsing without loading the entire dataset into memory, critical for training pipelines on resource-constrained hardware. The jsonlines structure allows metadata (e.g., source subset, document ID) to be stored alongside text.

Explicitly enumerates the 22 constituent subsets of the Pile (academic papers from PubMed and ArXiv, books from Books3 and Gutenberg, code from GitHub, web text from OpenWebText2 and Pile-CC, specialized sources like USPTO patents, Ubuntu IRC, and Stack Exchange) and provides source attribution for each document. This transparency enables users to understand the composition of their training data, audit for potential biases or contamination, and selectively exclude subsets if needed. However, exact composition percentages and subset enumeration are not fully documented.

Includes curated subsets of academic papers (PubMed, ArXiv), specialized technical sources (USPTO patents, Stack Exchange), and code repositories (GitHub), providing dense coverage of high-signal, domain-specific text that is underrepresented in web-only corpora. These subsets are integrated into the broader corpus at a fixed ratio, ensuring that models trained on the Pile develop specialized knowledge in these domains without requiring separate fine-tuning. The inclusion of academic papers and code is particularly valuable for training models intended for scientific or technical applications.

Incorporates two book-focused subsets (Books3 and Gutenberg) providing long-form, narrative text with complex linguistic structures, enabling models to develop strong performance on coherent, multi-paragraph generation and understanding of narrative arcs. Books represent a fundamentally different text distribution than web text (longer documents, more complex grammar, narrative structure) and are valuable for training models intended for creative writing, summarization, or long-context understanding. The inclusion of both contemporary books (Books3) and public-domain classics (Gutenberg) provides temporal and stylistic diversity.

Combines two web-derived subsets (OpenWebText2 and Pile-CC) providing broad coverage of diverse web text while applying quality filtering and deduplication to reduce noise compared to raw Common Crawl. OpenWebText2 is derived from URLs shared on Reddit (a proxy for human-curated quality), while Pile-CC is a filtered subset of Common Crawl. Together, these subsets provide web-scale coverage without the extreme noise and duplication of raw web scrapes, balancing breadth with quality.

Provides a fixed, immutable 825 GiB snapshot of the Pile corpus, enabling reproducible model training and evaluation across teams and time periods. The static nature ensures that models trained on the Pile in 2021 can be compared directly with models trained in 2024 without worrying about dataset drift or updates. However, no explicit versioning scheme, release notes, or update mechanism is documented, limiting transparency about potential corrections or improvements.

Provides formal citation guidance (Gao et al., 2020, arXiv:2101.00027) for the Pile itself and requires attribution to individual component datasets, establishing a precedent for proper data provenance documentation in large pretraining corpora. This framework enables researchers to trace the lineage of their training data and acknowledge the original sources and curators. However, no machine-readable citation metadata or automated attribution tools are provided.

Enables reproducible, open-source language model training by providing a publicly-available, freely-downloadable dataset used to train GPT-NeoX, Pythia, and other open models. The dataset is released under an open license (exact license terms not specified in artifact), allowing researchers and organizations to train models with full transparency and reproducibility. The Pile has influenced the design of subsequent open datasets, establishing a standard for open-source LLM training data.