I think they are using llama.cpp without any NPU/TPU patches. By default it only runs on CPU with support for various SIMD extensions. https://github.com/ggerg

I think they are using llama.cpp without any NPU/TPU patches. By default it only runs on CPU with support for various SIMD extensions. https://github.com/ggerganov/llama.cpp https://github.com/ggerganov/llama.cpp