Hugging Face has introduced a new feature that allows users to easily build and share kernels for the ROCm platform. This means developers can create, test, and deploy optimized code for AMD GPUs using a simple interface. The company claims this will make it easier for researchers and engineers to work with ROCm, which is an open-source platform for high-performance computing. Hugging Face's solution uses its popular Transformers library to provide a user-friendly way to build kernels without requiring extensive knowledge of low-level programming languages. --- Why it matters: This matters because it simplifies the process of working with ROCm, making it more accessible to researchers and engineers who want to leverage AMD GPUs for their projects. This can lead to faster development and deployment of AI models on these devices. Source: https://huggingface.co/blog/build-rocm-kernels