luban-cli_skill

Overview

This skill describes the Luban CLI for MLOps, a command-line tool to manage experiment environments, training jobs, and online services. It helps teams provision reproducible dev workspaces, orchestrate model training, and deploy scalable inference services. The content focuses on practical development patterns, command examples, and lifecycle operations for each entity.

How this skill works

The Luban CLI exposes three main command groups: env (environments), job (training tasks), and svc (online services). Each group implements CRUD operations so you can create, list, update, and delete resources. Developers extend the CLI using a Python boilerplate and follow the MLOps guide to keep commands consistent and automatable.

When to use it

• When you need repeatable dev environments for experiments and collaboration
• When orchestrating and monitoring training workloads across machines or clusters
• When deploying, scaling, or rolling back inference services
• When building or extending a Python-based MLOps CLI for automation
• When standardizing lifecycle operations across env, job, and svc entities

Best practices

• Start new commands from the provided Python boilerplate to maintain consistent CLI structure
• Implement full CRUD for every entity to support the entire resource lifecycle
• Use clear, consistent flags and subcommands following the referenced MLOps guide
• Include status and logging outputs for jobs and services to simplify debugging
• Keep environment images and job scripts versioned to ensure reproducibility

Example use cases

• Provision a reproducible research environment: luban env create --name research-v1 --image pytorch:2.0
• Run and track a training run: luban job create --script train.py --gpu 1 then luban job status --id job_001
• Deploy a model and scale replicas: luban svc create --model-path ./models/v1 --replicas 3 and luban svc scale --id my-service --replicas 5
• Extend the CLI with a new job subcommand using templates/cli_boilerplate.py
• Automate cleanup in CI by scripting luban env delete and luban job delete after tests