Blueprint by nvidia

Run models with llama.cpp on DGX Spark

Build llama.cpp with CUDA and serve models via an OpenAI-compatible API