DVC — data & model versioning¶
Git tracks code beautifully and large binaries terribly: a 2 GB dataset or a 500 MB model checkpoint bloats history irreversibly, because git's delta-compression assumes text-like files whose diffs are small — a retrained model is a different file from byte zero, so git stores a brand-new full copy on every commit. DVC (Data Version Control) fixes this by adding "git for data and models": it versions large files by storing a lightweight pointer in git (a path plus a content hash) while the actual bytes live in a remote (S3, GCS, Azure Blob, SSH, or a shared NAS), and it turns your training steps into a reproducible pipeline (a DAG of stages with cached outputs). …
🔒 La suite est en accès freemium — lecture complète 100 % gratuite
Tu lis ici l'aperçu libre. Le reste du chapitre (code, schémas, maths, exercices) fait partie du livre complet : crée un compte gratuit (30 secondes, aucun paiement) pour tout lire.