Skip to content

CLI Command Reference

This document provides a comprehensive reference for the command-line interfaces used in the Edge Computing LLM platform.

edge-cli

The edge-cli tool manages the lifecycle of the LLM platform on the edge node.

Global Flags

  • --kubeconfig: Path to the kubeconfig file (default: ~/.kube/config).
  • --debug: Enable verbose debug logging.
  • --output (-o): Output format (text, json, yaml).

Commands

edge-cli doctor

Runs pre-flight checks to ensure the system meets prerequisites (GPU drivers, memory, K3s installation). 

edge-cli doctor

edge-cli apply

Deploys or updates the platform components based on a specified profile. - --profile: The deployment profile (lite, standard, heavy). Default: standard. - --dry-run: Simulate the deployment without making changes.

edge-cli apply --profile standard

edge-cli verify

Validates the operational state of the deployed platform (e.g., checking if Ollama can access the GPU). 

edge-cli verify

edge-cli logs

Collects and aggregates logs from platform components. - --namespace (-n): Specific namespace to pull logs from. - --tail: Number of lines to retrieve.

edge-cli logs --namespace llm-observability --tail 100

edge-cli uninstall

Removes the platform components from the cluster. - --purge: Also delete PersistentVolumeClaims (PVCs) and downloaded models.

k3s-nvidia-edge

Used for low-level cluster provisioning and NVIDIA runtime configuration.

  • k3s-nvidia-edge install: Installs K3s tailored for NVIDIA GPU passthrough.
  • k3s-nvidia-edge configure-runtime: Sets up the NVIDIA Container Toolkit and configures containerd.

edge-llm-tests

The test execution wrapper script.

  • --mode: repository, cluster, or all.
  • --suite: Specific test suite to run (e.g., gpu, models, network).
  • --profile: Specify the profile context for tests (lite, standard, heavy).
./run-tests.sh --mode cluster --suite gpu

gguf-observe (Optional CLI utility)

A utility for interrogating the local observability stack directly.

  • gguf-observe metrics: Dumps current PromQL metrics for the GPU.
  • gguf-observe dashboard export: Exports current Grafana dashboard state.

edge-openai

Used to interact with the LLM via an OpenAI-compatible interface for testing.

  • edge-openai verify: Runs a suite of benchmark prompts against the local API to verify latency and correctness.
  • --endpoint: Override the default API endpoint.
  • --model: Specify the model to target.
edge-openai verify --model llama3:8b