Ollama implements cross-platform inference under the hood via llama.cpp, which was developed by Georgi Gerganov and is a pure C/C++ large model inference engine.