Hugging Face hosts 568 plug-and-play compute kernels, filterable by 200+ hardware platforms
ariG23498 · x · 2026-10-07
Hugging Face's Kernels hub now lists 568 compute kernels that can be browsed like models and datasets, filtered by hardware (H100, A100, Apple M4, WebGPU, and 200+ more) and plugged directly into your model. Popular entries include flash-attn3 (28.9k), flash-attn2, deep-gemm, and ggml quantization/attention kernels across CUDA, ROCm, and Metal.
More from coding & agent
- DIY "slop cannon" agent pipeline generates content for $0.38 per piece — tobowers · 2026-10-07
- Open-source jiti-lfe turns chat into a persistent LFE kernel for building live apps — arthurcolle · 2026-10-07
- Should AI agents have their own bank accounts and messaging identities? — amplifiedamp · 2026-10-07
- Dev builds persistent AI village with six Jev-driven characters to test personality consistency — Zazzen · 2026-10-07
- Dev uses vmpal agent to install Windows 11 in a VM, reacts with mock horror — DanielLockyer · 2026-10-07
- Free agent memory hub bundles 11 tutorials, from 7-minute explainer to offline edge RAG — Al_Grigor · 2026-10-07