Finally, some usable AI acceleration on the Linux side. All these hardware NPU investments will be wasted if they can't be used to power real world apps like LLMs