Show HN: Xet Server – Host your own models anywhere (github.com)

A new project called Xet Server has been introduced, enabling users to host their own AI models using a Go server compatible with Hugging Face's storage protocol. This innovative solution employs a content-defined chunking and deduplication storage system, replacing Git LFS for large model files. Users can seamlessly upload and download models via pre-existing Hugging Face CLI commands, benefiting from features such as byte-identical round-trips, BLAKE3 hashing, and LZ4 compression. The significance of Xet Server lies in its potential to enhance the resilience and efficiency of AI model storage and retrieval, particularly for developers facing internet connectivity issues. Notably, the inclusion of a caching proxy (xet-proxyd) allows for offline model access even if the Hugging Face Hub becomes unreachable. With zero external dependencies, pluggable storage options, and advanced file integrity checks, Xet Server presents a robust, user-friendly alternative for managing AI models, boosting both accessibility and reliability within the AI/ML community.