Zml-smi: universal monitoring tool for GPUs, TPUs and NPUs (zml.ai) AI

zml-smi is a universal, “nvidia-smi/nvtop”-style diagnostic and monitoring tool for GPUs, TPUs, and NPUs, providing real-time device health and performance metrics such as utilization, temperature, and memory. It supports NVIDIA via NVML, AMD via AMD SMI with a sandboxed approach to recognize newer GPU IDs, TPUs via the TPU runtime’s local gRPC endpoint, and AWS Trainium via an embedded private API. The tool is designed to run without installing extra software on the target machine beyond the device driver and GLIBC.