skills.rest
Mang30
Mang30COMMUNITY
Explore all skills by this author →
GitHub

run-experiment

Launches ML training jobs on local, SSH, Vast.ai, or Modal GPUs with logging.

Updated May 29, 2026
Search Tags:#modal#wandb#vast.ai#ml experiments#training jobs#gpu provisioning#ssh deployment
One-click install
npx skills add https://github.com/Mang30/myskills --skill run-experiment-mang30
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: run-experiment
Source: https://github.com/Mang30/myskills/tree/main/skills/run-experiment
Command: npx skills add https://github.com/Mang30/myskills --skill run-experiment-mang30

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It eliminates the hassle of manually provisioning GPUs and launching training jobs by turning an experiment request into a repeatable deploy-and-run workflow across local machines, SSH servers, Vast.ai, or Modal.

Core Features & Use Cases

  • Environment-aware deployment: Detects whether the target GPU is local, remote (SSH), Vast.ai, or Modal using the project’s CLAUDE.md settings.
  • GPU pre-flight validation: Checks GPU availability (e.g., free memory threshold) before launching to avoid wasting time on unsuitable hardware.
  • Job launch with logging: Starts the training in a dedicated screen session (remote/Vast) or a background/local process, capturing logs with tee for later review.
  • Platform-specific lifecycle handling: Uses Vast.ai instance reuse or provisioning, and can auto-destroy Vast instances after completion to prevent idle billing.
  • Optional W&B automation: When enabled in CLAUDE.md, adds/ensures Weights & Biases logging and verifies wandb login on the target machine.
  • Use case: You need to run a new training run for a model update and want it executed on the best available GPU with an estimated cost and saved logs, regardless of where that GPU lives.

Quick Start

Tell the AI to run an experiment with your training entrypoint and arguments, and it will detect the environment from CLAUDE.md, select or provision the right GPU, launch the job, and return the running details.

Frequently Asked Questions about run-experiment

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

FAQPage Schema
How do I automatically provision and run machine learning experiments across different GPUs?▼

To run machine learning experiments automatically, this Skill detects your target compute environment from CLAUDE.md metadata and launches training jobs with logging across local GPUs, SSH servers, Vast.ai, or Modal.

What's the best way to manage Vast.ai instance provisioning and avoid idle billing?▼

To manage Vast.ai instances and avoid idle billing, this Skill provisions or reuses compute instances and can auto-destroy them after training completion, preventing ongoing charges for unused hardware.

Do I need a CLAUDE.md file to route ML training jobs to remote SSH servers or Modal?▼

Yes, a CLAUDE.md file with environment metadata is required for GPU routing. This Skill uses it to detect whether to deploy via local execution, remote SSH servers, Vast.ai, or Modal serverless GPUs.

Can I automatically inject Weights and Biases logging into my training jobs?▼

Yes, you can inject Weights and Biases logging. When W&B is enabled in CLAUDE.md, this Skill adds the necessary logging code and verifies wandb login on the target machine before starting the job.

How does GPU pre-flight validation work before launching training jobs?▼

GPU pre-flight validation checks GPU availability, including free memory thresholds, before launching training jobs to ensure the hardware is suitable and avoid wasting time on unsuitable compute resources.

Does Modal serverless GPU execution require a manual availability check?▼

No, Modal serverless GPU execution does not require a manual availability check. The Skill skips GPU pre-flight validation for Modal, unlike local, SSH, and Vast.ai environments where checks are mandatory.

Related Skills

Explore More Education & Research Skills→
Leo1349Leo1349

run-experiment

Deploy machine learning experiments to local or remote GPU servers via SSH.

Community
Advanced
LingrongyeLingrongye
1

monitor-experiment

Monitor running experiments via SSH and collect JSON results.

Community
Intermediate
raja21068raja21068
2

run-experiment

Deploy and run machine learning experiments across local GPUs, SSH servers, Vast.ai, and Modal.

Community
Advanced
randrand
126

discover-ml

Discover ML and AI skills during model training and inference workflows.

Community
Basic