TechPrufer
Get started
Hugging Face
Hugging Face logoHugging Face · Feature trail

Custom Kernel Distribution and Security

1 post · Jan 2026 – Jul 2026

What this trail covers

Hugging Face has significantly enhanced its 🤗 Kernels project, introducing a new 'kernel' repository type on the Hub for better discoverability and management of custom compute kernels. Security has been bolstered with trusted kernel publishers, kernel signing using Sigstore cosign, and reproducible builds via Nix. The CLIs for `kernels` and `kernel-builder` have been revamped for better separation of concerns and agent optimization. Support for frameworks like Torch Stable ABI and Apache TVM F. This post details how Claude was used to generate CUDA kernels for diffusion models, demonstrating an AI-assisted approach to kernel development and distribution, and how these kernels can be used to teach open models.

All Hugging Face trails
  1. 🤗 Kernels: Major UpdatesThis post details major updates to the 🤗 Kernels project, including the introduction of a new 'kernel' repository type on the Hugging Face Hub for specialized compute kernels. Security enhancements include the introduction of 'trusted kernel publishers' and kernel signing using Sigstore cosign, with reproducible builds enforced via Nix. The CLIs for `kernels` and `kernel-builder` have been refactored for clearer separation of concerns and agent optimization. Framework support has been extended to include Torch Stable ABI and Apache TVM FFI. The project now provides a foundation for agentic kernel development, enabling agents to scaffold, build, benchmark, and iteratively optimize kernels through structured workflows and integrated benchmarking with HF Jobs.
  2. We Got Claude to Build CUDA Kernels and teach open models!This post details the use of Claude, an AI model, to generate CUDA kernels for diffusion models. It showcases an agentic workflow where an AI model assists in the development and distribution of custom compute kernels. The process involved using Claude to write CUDA code, which was then integrated into the Hugging Face kernel distribution system. This demonstrates a novel approach to kernel development and highlights the potential of AI in accelerating the creation of specialized compute components for ML workloads.