Hugging Face officially launches Kernels, GPU operators ready to use with just one line of code like a model

ME News Report, April 15 (UTC+8), according to 1M AI News monitoring, Hugging Face CEO Clem Delangue announced that Kernels is officially launched on the Hub. GPU operators are low-level optimized codes that push graphics cards to their limits, capable of accelerating inference and training by 1.7 to 2.5 times, but installation has always been a nightmare: taking FlashAttention as the most common example, local compilation requires about 96GB of memory and several hours; even slight mismatches in PyTorch or CUDA versions cause errors, causing most developers to get stuck at this step. Kernels Hub moves compilation to the cloud. Hugging Face pre-compiles operators across various graphics cards and system environments; developers write a single line of code, and the Hub automatically matches the hardware environment, downloading pre-compiled files that are ready to use within seconds. Multiple different versions of operators can be loaded in the same process, compatible with torch.compile. Kernels was tested and launched in June last year, and this month was upgraded to a first-level repository on the Hub, alongside Models, Datasets, and Spaces. Currently, there are 61 pre-compiled operators covering common scenarios such as attention mechanisms, normalization, mixture-of-experts routing, and quantization, supporting four hardware acceleration platforms: NVIDIA CUDA, AMD ROCm, Apple Metal, and Intel XPU. It has been integrated into Hugging Face’s inference framework TGI and Transformers library. (Source: BlockBeats)