JFrog Artifactory now natively supports the Xet protocol developed by Hugging Face, enabling chunk-level deduplication and up to 3.7x faster model downloads. Xet splits large binary files into content-defined chunks stored in 64 MB 'xorb' bundles, so only genuinely new data is stored — fine-tuning a 50 GB model that changes 2% of weights transfers only the changed chunks. Artifactory extends its existing checksum-based deduplication engine to Hugging Face repositories, caches xorb ranges shaped by real usage, and applies intelligent rate limiting to prevent upstream 429 errors. Additional enterprise benefits include production continuity independent of upstream availability, centralized access management for gated models like Llama and Gemma, and security scanning of incoming models. Xet support is currently available for Hugging Face remote repositories and can be enabled via a single checkbox in repository settings.

9m read timeFrom jfrog.com
Post cover image
Table of contents
Why Xet?How does JFrog support Xet and Hugging Face?Which Artifactory repositories support the Xet protocol?How can I turn on Xet support for my Hugging Face Repositories?How does Xet support work for JFrog Artifactory under the hood?Why JFrog is the best option for enterprise model management?Get started with Xet and JFrog Artifactory today
16 Impressions