#Machine Learning
Machine learning hardware on Linux: CUDA containers on an NVIDIA DGX H100 with RHEL and Google Coral Edge TPUs on Fedora.
- Swapping LLMs in Under Two Seconds: A Sleep Mode for llama.cpp
How a small wrapper that parks llama-server's GPU memory in host RAM, plus a kernel setting, makes model swaps behind llama-swap about six times faster.
- RHEL 8 CUDA Container Install on a DGX H100 Machine
Installing the NVIDIA driver, CUDA, Fabric Manager and the container toolkit on an NVIDIA DGX H100 running RHEL 8.8.
- Installing Google Coral Mini PCIE Dual EdgeTPU on Fedora
Running a Google Coral Dual Edge TPU on Fedora: a PCIe adapter for both TPU cores and the gasket-dkms driver from Copr.