skills.rest
Lingjie-wang
Lingjie-wangCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

run-experiment

Deploy and run ML training experiments on local, remote, Vast.ai, or Modal GPU environments.

Updated Jul 9, 2026
Search Tags:#training#ssh#modal#wandb#vast.ai#gpu deployment#ml experiment
One-click install
npx skills add https://github.com/Lingjie-wang/autoRL --skill run-experiment-lingjie-wang
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: run-experiment
Source: https://github.com/Lingjie-wang/autoRL/tree/main/Auto-claude-code-research-in-sleep/skills/run-experiment
Command: npx skills add https://github.com/Lingjie-wang/autoRL --skill run-experiment-lingjie-wang

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Launching ML training jobs across heterogeneous GPU environments (local machines, SSH servers, rented Vast.ai instances, Modal serverless) requires repetitive manual work: checking GPU availability, syncing code, installing dependencies, starting screen sessions, and tracking costs. This Skill automates that entire deployment workflow from a single request. ## Core Features & Use Cases - Multi-Environment Deployment: Detects the target environment from the project's CLAUDE.md and deploys to local CUDA/MPS, remote SSH servers, Vast.ai instances, or Modal serverless GPUs. - Automated Code Sync & Setup: Syncs code via rsync or git, installs dependencies in ordered phases, and optionally injects W&B logging into training scripts. - Full Lifecycle Management: Launches experiments in dedicated screen sessions with GPU binding, verifies launch, sends Feishu notifications, and auto-destroys Vast.ai instances with cost reporting. - Use Case: A researcher says "run the fine-tuning experiment on a rented GPU" — the Skill provisions a cost-optimized Vast.ai instance, syncs the code, launches training with W&B logging, downloads results, and destroys the instance to stop billing. ## Quick Start Ask the assistant to run your training script as an experiment, for example: run the train.py experiment on a remote GPU with W&B logging enabled.

Frequently Asked Questions about run-experiment

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run an ML training experiment on a remote GPU server?▼

Add your server details (SSH alias, conda environment, code directory) to the project's CLAUDE.md with gpu: remote, then request the experiment run. The Skill checks GPU availability with nvidia-smi, syncs code via rsync, and launches training in a dedicated screen session.

How to run experiments on Vast.ai rented GPUs automatically?▼

Set gpu: vast in CLAUDE.md after installing the vastai CLI and setting your API key. The Skill provisions a cost-optimized instance, syncs code to /workspace/project/, runs the experiment, downloads results, and auto-destroys the instance to stop billing.

What is the difference between Vast.ai and Modal for running experiments?▼

Vast.ai rents dedicated GPU instances requiring SSH, code sync, and manual or automatic destruction. Modal is serverless with no SSH or code sync — it handles GPU allocation automatically, scales to zero when idle, and returns results via Volumes.

Does the experiment runner support Weights & Biases logging?▼

Yes, set wandb: true with wandb_project in CLAUDE.md. The Skill checks training scripts for existing wandb integration and automatically adds wandb.init and wandb.log calls for training loss, learning rate, eval metrics, and GPU memory if absent.

Can I run multiple experiments in parallel on different GPUs?▼

Yes, each experiment gets its own screen session bound to a specific GPU via CUDA_VISIBLE_DEVICES. The Skill checks GPU memory first (free means under 500 MiB used) and launches multiple experiments in parallel on different GPUs.

Why does the Skill check GPU memory before launching training?▼

It runs nvidia-smi to verify memory.used is below 500 MiB, preventing job failures from assigning experiments to occupied GPUs. This pre-flight check applies to local, SSH, and Vast.ai targets, while Modal manages allocation automatically.

Related Skills

Explore More Education & Research Skills→
tqLi99tqLi99

run-experiment

Deploy and run ML experiments on local or remote GPU environments via SSH.

Community
Advanced
duypham2801duypham2801

monitor-experiment

Monitor running ML experiments and collect logs, metrics, and JSON results.

Community
Intermediate
zhuyingqinzhuyingqin
1

vast-gpu

Provision and manage vast.ai GPU instances for training tasks.

Community
Advanced
auxi-wardrobeauxi-wardrobe
1

ck:devops

Automate cloud infrastructure deployment across Cloudflare, Docker, Google Cloud, and Kubernetes.

Official
Advanced