Hugging Face Kernels: swap kernels via mapping changes, no model rewrite needed
RisingSayak · x · 2026-09-18
- Hugging Face's Kernels library lets you swap in compatible kernels for existing layers as a mapping change, not a model rewrite — try one, benchmark your workload, keep what helps.
- You can specify separate kernels for inference vs. training, and different kernels per device type, registered globally or locally in a mapping.
- Mechanism: annotate a layer with usekernelforwardfromhub() (e.g. SiluAndMul); kernelize() looks up Hub kernels by name to replace the forward method.
Related event: Hugging Face Launches Kernels Library for Drop-in Faster Kernels(2 posts)→
More from Infra
- Anthropic Targets ~10GW Compute by 2027, in Talks with Meta on $10B Lease — FlorianGallwitz · 2026-09-18
- Shield AI exec lays out dual strategy: multi-model intelligence and sovereign compute — BrettKrieger12 · 2026-09-18
- Cloudflare is building a new CI/CD system you customize with TypeScript — dinasaur_404 · 2026-09-18
- huggingface_hub v1.32 dedupes Xet files in local cache: 5 repos sharing 20GB weights now use ~20GB — victormustar · 2026-09-18
- Apple reportedly mulls M8 Ultra-based AI enterprise servers with NVLink Fusion, eyed for 2029 — Beth_Kindig · 2026-09-18
- ProgramAsWeights: compile English-described AI functions into tiny CPU-only neural nets — yuntiandeng · 2026-09-18