What to install - Warpscale

Choose what you need

The probe runs on every GPU node and is required for everything else: the CLI and the vLLM plugin report through it, and the Python SDK reports through the CLI. What you add alongside the probe depends on your workload.

Component	Install	When you need it
Probe	Install the probe	Always. One per GPU node.
CLI	Install the CLI	Training runs.
Python SDK	Install the Python SDK	Training runs. Adds step, epoch, and checkpoint progress from inside your code.
vLLM plugin	Install the vLLM plugin	Inference on vLLM.

Requirements

These apply to every component and every deployment mode. Each install page lists what it needs on top of these. Platform

Linux only.
Architecture amd64 or arm64.
glibc 2.31 or newer — Ubuntu 20.04+, Debian 11+, RHEL 9+.
cgroup v2, and a kernel with eBPF and BTF enabled.
An NVIDIA driver installed and reachable by the probe, for GPU metrics.

Account

Your organization URL
An organization API key

Both are set up when your organization is created. See the Quickstart.