MINT-1T-PDF-CC-2023-40 vs The Stack v2

Extracts text content from 1 trillion tokens of PDF documents using OCR and layout-aware parsing, preserving document structure and spatial relationships. The dataset combines Common Crawl PDF snapshots with machine-readable text extraction, enabling training of models that understand both visual layout and semantic content. Architecture uses distributed PDF processing pipelines to handle heterogeneous document formats (scanned PDFs, native PDFs, mixed content) across 857K+ document samples.

Unique: Combines 1 trillion tokens of Common Crawl PDFs with layout-aware extraction preserving spatial document structure, unlike generic text corpora that discard formatting. Uses distributed PDF parsing to handle heterogeneous document types (scanned, native, mixed) at web scale rather than curated document collections.

vs alternatives: Larger and more diverse than academic document datasets (e.g., DocVQA, RVL-CDIP) while maintaining layout information that generic text corpora like C4 or The Pile discard entirely.

Provides structured image-text pairs extracted from PDF documents where images are document pages and text is extracted content, enabling direct training of vision-language models without manual annotation. The dataset architecture preserves the natural alignment between visual document layout and corresponding text, creating implicit supervision signals. Processing pipeline handles page segmentation, text-image alignment, and quality filtering across millions of document samples.

Unique: Leverages natural document structure to create implicit image-text alignment without manual annotation, using page-level visual-semantic correspondence from PDFs. Unlike manually-annotated datasets (Flickr30K, COCO), derives pairs automatically from document layout, enabling trillion-token scale.

vs alternatives: Provides orders of magnitude more image-text pairs than manually-curated datasets while maintaining document-specific semantic alignment that generic web image-text pairs (Laion) lack.

Supplies 1 trillion tokens of English text extracted from PDF documents, suitable for pretraining or continued training of large language models. The corpus is derived from diverse document sources across Common Crawl, providing varied writing styles, domains, and content types. Processing pipeline includes tokenization, deduplication, and quality filtering to ensure training data suitability while maintaining scale.

Unique: Derives 1 trillion tokens specifically from PDF documents rather than generic web crawls, capturing formal, structured writing with higher information density than typical web text. Preserves document-level context and structure signals that web-only corpora lose.

vs alternatives: Complements web-text corpora (C4, The Pile) by providing document-sourced content with different statistical properties, useful for models requiring strong document understanding capabilities.

Enables selective access to dataset subsets filtered by document characteristics (source domain, document type, quality metrics) without downloading the full 1 trillion token corpus. The dataset infrastructure supports streaming access with client-side filtering, allowing researchers to construct domain-specific training sets from the larger collection. Filtering operates on document metadata including source URLs, extraction quality scores, and document type classifications.

Unique: Provides streaming access with metadata-based filtering on trillion-token dataset without requiring full download, using Hugging Face Datasets infrastructure for efficient subset construction. Enables on-demand domain-specific corpus creation from larger collection.

vs alternatives: More flexible than fixed-size domain datasets (e.g., ArXiv papers, legal documents) by allowing dynamic filtering from larger corpus; more efficient than downloading full dataset for subset access.

Maintains document layout information (page structure, text positioning, formatting) during PDF-to-text conversion, enabling models to learn relationships between visual layout and semantic content. The extraction pipeline preserves spatial coordinates, text ordering, and structural hierarchy (headings, sections, lists) rather than flattening documents to linear text. This architectural choice enables training of layout-aware models that can reason about document organization.

Unique: Preserves document layout and spatial relationships during extraction rather than flattening to linear text, enabling training of models that understand how document organization conveys meaning. Uses coordinate-aware parsing to maintain structural hierarchy.

vs alternatives: Enables layout-aware training unlike text-only corpora (C4, The Pile) while providing larger scale than manually-annotated layout datasets (DocVQA, RVL-CDIP).

Provides access to a specific snapshot of PDF documents from Common Crawl (2023-40 version), with consistent versioning and reproducibility guarantees. The dataset is built from a fixed Common Crawl snapshot, enabling reproducible research and consistent data across training runs. Infrastructure includes metadata linking documents to their Common Crawl source, enabling traceability and potential re-extraction with updated pipelines.

Unique: Provides versioned, reproducible access to specific Common Crawl PDF snapshot (2023-40) with full provenance tracking, enabling research reproducibility. Unlike generic Common Crawl access, includes pre-processed extraction and structured metadata.

vs alternatives: More reproducible than direct Common Crawl access (which changes over time) while providing pre-processed documents unlike raw Common Crawl snapshots.

Aggregates 67 TB of source code from the Software Heritage archive, filtering for permissively licensed repositories (MIT, Apache 2.0, BSD, etc.) across 600+ programming languages. Uses automated license detection and validation to ensure legal compliance for model training. Implements a rigorous deduplication pipeline at file and repository levels to eliminate redundant training data and reduce dataset bloat.

Unique: Largest open-source code dataset at 67 TB with automated opt-out governance allowing repository owners to request removal, combined with rigorous deduplication and PII removal pipeline — no other public dataset offers this scale with legal compliance and community control mechanisms

vs alternatives: Larger and more legally compliant than GitHub's CodeSearchNet (14M files) or Google's BigQuery public datasets, with explicit opt-out governance vs. implicit inclusion, and covers 600+ languages vs. Codex training data's undisclosed language distribution

Implements a community-driven opt-out system where repository owners can request removal of their code from the dataset without legal takedown notices. Maintains a registry of excluded repositories and re-applies exclusions during dataset updates. Provides transparent governance documentation and a clear submission process for removal requests, balancing open access with creator rights.

Unique: First large-scale code dataset to implement opt-out governance at dataset level rather than relying solely on license compliance, with transparent registry and community submission process — shifts power from dataset creators to code contributors

vs alternatives: More respectful of creator autonomy than GitHub Copilot's training approach (no opt-out) or academic datasets (one-time snapshot), and more scalable than individual DMCA takedowns

Automated pipeline that scans source code for personally identifiable information (email addresses, API keys, SSH keys, credit card patterns, phone numbers) and removes or redacts them before dataset release. Uses regex patterns, entropy-based detection for secrets, and heuristic rules to identify sensitive data. Operates at file level with configurable sensitivity thresholds to balance data utility against privacy risk.

Unique: Combines regex pattern matching, entropy-based secret detection, and heuristic rules in a unified pipeline with configurable sensitivity — more comprehensive than simple regex-only approaches, but trades off false positive rate against security coverage

vs alternatives: More thorough than GitHub's secret scanning (which only flags known patterns) because it includes entropy-based detection for unknown secret formats, but less accurate than specialized tools like TruffleHog due to language-agnostic approach

Indexes 67 TB of source code across 600+ programming languages with language-aware metadata (syntax, file extension, language family). Enables retrieval by language, license, repository, or code patterns. Uses Software Heritage's existing indexing infrastructure as foundation, augmented with language detection and classification. Supports both bulk download and filtered queries for specific language subsets.

Unique: Leverages Software Heritage's existing language detection and indexing infrastructure, then augments with BigCode-specific language classification and filtering — avoids reinventing language detection while providing dataset-specific query capabilities

vs alternatives: More comprehensive language coverage (600+ languages) than GitHub's Linguist (500+ languages) and more accessible than Software Heritage's raw API because it's pre-filtered for permissive licenses and deduplicated

Removes duplicate code files and repositories using content hashing (SHA-256 or similar) and fuzzy matching for near-duplicates. Operates in two stages: exact deduplication via hash matching, then fuzzy matching (e.g., Jaccard similarity or MinHash) to catch semantically identical code with minor formatting differences. Preserves one canonical copy of each unique code pattern while removing redundant training examples.

Unique: Two-stage deduplication combining exact hash matching with fuzzy similarity matching (likely MinHash or Jaccard) to catch both identical and near-identical code — more thorough than single-stage approaches but computationally expensive

vs alternatives: More aggressive deduplication than CodeSearchNet (which uses simple hash matching) because it catches near-duplicates, but less semantic than clone detection tools (which understand code structure) because it's content-based

Integrates with Software Heritage's comprehensive archive of 200+ million repositories and their full version control history. Extracts source code snapshots from Software Heritage's Git/Mercurial/SVN repositories, preserving repository metadata (commit history, author info, timestamps). Provides access to code at specific points in time, enabling historical analysis or training on code evolution patterns.

Unique: Leverages Software Heritage's universal code archive (200M+ repositories) as data source, providing access to code that would be impossible to collect via GitHub API alone — enables training on archived/deleted repositories and non-GitHub platforms (GitLab, Gitea, etc.)

vs alternatives: More comprehensive than GitHub-only datasets because it includes code from GitLab, Gitea, SourceForge, and other platforms archived by Software Heritage; more legally defensible than web scraping because it uses an established, community-maintained archive

Tracks and validates SPDX license identifiers for each repository, ensuring only permissively licensed code (MIT, Apache 2.0, BSD, etc.) is included. Maintains license metadata alongside code files, enabling downstream users to verify legal compliance. Implements license hierarchy and compatibility checking to handle dual-licensed or complex licensing scenarios.

Unique: Combines automated SPDX detection with manual review and maintains license metadata alongside code, enabling downstream users to verify compliance — more transparent than datasets that simply claim 'permissive licenses' without proof

vs alternatives: More legally rigorous than GitHub's CodeSearchNet (which doesn't validate licenses) and more transparent than Codex training data (which doesn't disclose license filtering at all)

Maintains versioned snapshots of the dataset (e.g., v2.0, v2.1) with documented changes between versions (new repositories added, deduplication improvements, PII removal updates). Provides checksums and manifests for reproducibility, enabling researchers to cite specific dataset versions and reproduce results. Tracks dataset lineage and transformation history.

Unique: Maintains semantic versioning and detailed changelogs for dataset releases, enabling researchers to cite specific versions and understand dataset evolution — more rigorous than one-off dataset releases without versioning

vs alternatives: More reproducible than academic datasets that are released once without versioning, and more transparent than commercial datasets (Codex) that don't disclose version history or changes