MINT-1T-PDF-CC-2023-06 vs The Stack v2

Provides a curated dataset of 1 trillion tokens spanning 539,406 PDF documents with aligned image-to-text pairs extracted from Common Crawl 2023-06 snapshot. The dataset uses a hierarchical indexing structure that maps document boundaries, page-level image coordinates, and corresponding OCR/text extractions, enabling efficient retrieval of multimodal training samples at scale without requiring full dataset materialization in memory.

Unique: Combines 1 trillion tokens of document text with aligned page-level images from a single Common Crawl snapshot, providing temporally-consistent multimodal pairs at unprecedented scale — most competing datasets either use synthetic image-text pairs or lack document-level coherence across modalities

vs alternatives: Larger and more document-focused than LAION-5B (which emphasizes web images) and more naturally-paired than synthetic datasets like Synthetic Docvqa, with real-world OCR challenges that improve model robustness

Implements HuggingFace Datasets streaming protocol that enables on-demand loading of document samples without downloading the full 1T token dataset upfront. The architecture uses memory-mapped file access and configurable batch sampling strategies, allowing training loops to fetch and cache only the samples needed for each epoch while maintaining deterministic shuffling across distributed workers.

Unique: Uses HuggingFace's streaming protocol with deterministic shuffling and worker-aware sharding, enabling true distributed training without pre-downloading — avoids the storage bottleneck that limits competitors like LAION-5B when used in multi-node setups

vs alternatives: More practical for large-scale training than downloading full datasets upfront, and more deterministic than ad-hoc web scraping approaches that lack reproducibility

Maintains structured metadata for each document including source URL, Common Crawl snapshot date (2023-06), document hash, page count, and extraction quality scores. This metadata is queryable and filterable within the dataset, allowing users to select subsets based on source domain, quality thresholds, or temporal characteristics without scanning the full corpus.

Unique: Embeds Common Crawl provenance (URLs, crawl dates, document hashes) directly in the dataset schema, enabling reproducible filtering and bias analysis — most competing datasets either lack this metadata or store it separately, making it harder to correlate quality with source

vs alternatives: Provides better auditability and reproducibility than datasets without source tracking, and more granular filtering than datasets with only aggregate statistics

Extracts page-level images from PDF documents and aligns them with corresponding OCR/text content using spatial layout information (bounding boxes, reading order). The extraction pipeline preserves document structure (headers, footers, tables, body text) by analyzing PDF internal structure and image coordinates, creating naturally-aligned multimodal pairs suitable for vision-language model training without requiring post-hoc alignment.

Unique: Preserves document layout structure through PDF internal coordinate systems rather than post-hoc image analysis, enabling structurally-aware alignment that captures reading order and spatial relationships — most competing datasets either discard layout information or infer it from image analysis alone

vs alternatives: More accurate layout alignment than image-only document datasets, and more scalable than manually-annotated document datasets like DocVQA

Dataset is derived from a single Common Crawl snapshot (2023-06), ensuring temporal consistency across all documents — all PDFs were crawled within a specific time window, avoiding temporal distribution shifts that occur when combining data from multiple crawl dates. The integration includes Common Crawl metadata (WARC records, crawl IDs) enabling users to trace documents back to original crawl artifacts for verification or re-extraction.

Unique: Anchors entire dataset to a single Common Crawl snapshot (2023-06) with traceable WARC references, ensuring temporal consistency and reproducibility — most competing web-derived datasets either combine multiple crawl dates or lack explicit Common Crawl integration

vs alternatives: More reproducible than datasets combining multiple crawl dates, and more verifiable than proprietary datasets without public provenance

Dataset is released under Creative Commons Attribution 4.0 (CC-BY-4.0) license, permitting commercial use, modification, and redistribution with attribution. The license is applied at the dataset level, though individual documents may have different licenses — users are responsible for verifying compliance for derived works, but the dataset itself imposes minimal legal restrictions on model training and deployment.

Unique: Explicitly licensed under CC-BY-4.0 with clear commercial use rights, reducing legal friction for commercial model training — many competing datasets either lack explicit licensing or use more restrictive licenses (e.g., non-commercial only)

vs alternatives: More commercially-friendly than datasets with non-commercial restrictions, and more legally transparent than datasets with unclear licensing

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