rocm-kernel
The inference engine the open-source world built for itself.
Zero-dependency LLM inference engine. Fastest cold start, smallest VRAM footprint. CUDA + ROCm + Metal.