Miner CLI Overview

This document describes the positioning, capabilities, and deployment file structure of miner-cli.

Positioning

miner-cli is a Docker-based deployment helper for running large language models on a single NVIDIA GPU Linux host.

It is intentionally narrow:

• No scheduler
• No cluster control plane
• No multi-tenant platform
• Config → Compose → container → health check only

Core Capabilities

miner-cli can:

• Check whether the machine is ready for GPU containers
• Generate a starter YAML config
• Render Docker Compose deployment files
• Start vllm or sglang containers
• Run GPU and runtime smoke tests
• Wait for /v1/models to become ready
• Manage logs, status, stop, restart, remove, and list
• Add dcgm-exporter and miner-agent sidecars when configured

Artifact Layout

Deployment files are rendered into:

~/.miner-cli/deployments/<name>/

File	Purpose
config.yaml	Copy of the source deployment config
compose.yaml	Rendered Docker Compose file
.env	Environment file containing configured Hugging Face token

Supported Engines

Currently supported:

• vllm (Current focus)
• sglang (Future support)

Image Tag Policy

The CLI supports stable and latest image policies when generating configs:

Engine	Default image
vllm	vllm/vllm-openai:latest

Production Recommendation

When a vllm image uses a floating latest tag, miner-cli warns during init, runtime prepare, and up. Production deployments should explicitly pin image.

Related Documentation

• miner-cli Commands
• miner-cli Configuration
• Quick Start