c4 vs The Stack v2

C4 ingests petabyte-scale Common Crawl snapshots and applies language detection, URL filtering, and exact/fuzzy deduplication to produce a cleaned multilingual corpus spanning 100+ languages. The pipeline uses probabilistic deduplication techniques and language-specific filtering rules to remove boilerplate, near-duplicates, and low-quality content while preserving linguistic diversity across 806 billion tokens.

Unique: C4 is built directly from Common Crawl snapshots with transparent, reproducible filtering and deduplication logic (published in the original paper), making it auditable and replicable — unlike proprietary datasets. It includes explicit language detection and URL-based quality filtering applied uniformly across 100+ languages, enabling fair multilingual representation.

vs alternatives: C4 offers 10x larger scale and true multilingual coverage compared to English-only datasets like Wikipedia or BookCorpus, while maintaining open-source transparency and reproducibility that proprietary datasets (e.g., GPT-3's training data) cannot provide.

C4 applies language-specific heuristics to filter low-quality documents, including URL-based blocklists (e.g., adult sites, spam domains), text quality metrics (line length, word count, symbol ratios), and language-specific stopword and boilerplate detection. Documents are ranked by quality signals and can be sampled probabilistically to balance dataset composition.

Unique: C4's filtering is fully transparent and reproducible — the exact rules, thresholds, and blocklists are published and can be audited or modified. This contrasts with proprietary datasets where filtering logic is opaque. The approach uses language-specific metrics rather than one-size-fits-all rules, acknowledging that quality signals differ across scripts and languages.

vs alternatives: C4's filtering is more transparent and auditable than proprietary datasets, while being simpler and more reproducible than learned quality models (which require labeled data and add complexity).

C4 applies two-stage deduplication: exact matching via SHA-256 hashing of normalized text, followed by fuzzy matching using MinHash sketches to identify near-duplicates with configurable Jaccard similarity thresholds. This removes redundant content while preserving legitimate repetition across the web, reducing dataset size by ~25% while maintaining diversity.

Unique: C4 combines exact and fuzzy deduplication in a two-stage pipeline, using MinHash for efficient approximate matching at scale. The approach is fully reproducible and the thresholds are published, allowing researchers to audit or adjust deduplication aggressiveness. This is more sophisticated than simple exact-match deduplication but simpler than learned semantic deduplication models.

vs alternatives: C4's two-stage deduplication is more scalable and transparent than semantic deduplication models, while catching more duplicates than exact-match-only approaches, making it practical for petabyte-scale datasets.

C4 detects document language using probabilistic language identification (langdetect library) and stratifies the corpus by language, enabling per-language filtering, quality ranking, and balanced sampling. The dataset supports 100+ languages with language-specific metadata, allowing users to select subsets by language or language family.

Unique: C4 provides explicit language detection and stratification for 100+ languages, enabling transparent per-language analysis and balanced sampling. This is more comprehensive than English-only datasets and more transparent than datasets with opaque language composition. The language metadata is included in the dataset, allowing users to audit and adjust language representation.

vs alternatives: C4's language detection and stratification enable true multilingual training and analysis, unlike English-only datasets, while maintaining transparency about language distribution and quality that proprietary multilingual datasets lack.

C4 is hosted on HuggingFace Hub and supports streaming access without downloading the full dataset, using the datasets library's streaming protocol. The dataset is partitioned into language and snapshot-specific shards, enabling distributed loading across multiple workers and machines. Users can load subsets by language, snapshot, or split without downloading the entire corpus.

Unique: C4 leverages HuggingFace Hub's streaming infrastructure to enable on-demand access without full downloads, using language and snapshot-based sharding for fine-grained parallelism. This is more practical than requiring users to download 750GB locally, and more flexible than static dataset snapshots.

vs alternatives: C4's streaming access via HuggingFace Hub is more practical than downloading the full dataset locally, while being more flexible and transparent than proprietary cloud-hosted datasets that require vendor lock-in.

C4 is built from specific Common Crawl snapshots (e.g., 2019-30, 2020-05) and maintains explicit versioning, allowing users to reproduce results with the exact same data. The dataset includes metadata about source snapshots, filtering parameters, and deduplication thresholds, enabling full lineage tracking and reproducibility of model training runs.

Unique: C4 provides explicit snapshot-based versioning tied to Common Crawl releases, with published filtering and deduplication parameters, enabling full reproducibility and lineage tracking. This is more transparent than datasets with opaque versioning or continuous updates that make reproduction difficult.

vs alternatives: C4's snapshot-based versioning enables reproducible research and auditable data sourcing, unlike continuously-updated datasets or proprietary datasets with opaque versioning.

C4 is built from Common Crawl (public domain) and applies URL-based filtering to exclude copyrighted content and adult sites, resulting in a corpus suitable for open-source model training without licensing restrictions. The dataset is released under the Open Data Commons Attribution License (ODC-BY), enabling commercial and research use with attribution.

Unique: C4 is explicitly designed for open-source model training, using Common Crawl (public domain) and applying URL-based filtering to exclude copyrighted content. The dataset is released under ODC-BY, enabling transparent, compliant use. This contrasts with proprietary datasets or datasets with unclear licensing.

vs alternatives: C4 provides a large, open-source corpus suitable for commercial model training, unlike proprietary datasets (which require licensing) or datasets with unclear legal status.

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