Skip to content
Get packs

Run ML Experiments on GPU Servers

Deploy and launch ML training jobs on local, SSH, Vast.ai, or Modal GPU machines.

Source Download skill
Repository
wanshuiyin/Auto-claude-code-research-in-sleep
Stars
18,482
Licence
MIT
Files
1 file

What it does

When you ask the agent to run a machine learning experiment, this skill tells it how to find your server or GPU setup from your project's AGENTS.md file, check that a GPU is free, copy only the needed code to the machine, and start the training script in a detached screen session with logging. It can also add Weights & Biases logging to a training script, send an optional Feishu notification, and shut down a Vast.ai cloud instance after results are saved. It works as a checklist so the agent does not launch jobs blindly.

Great for

  • ML researchers who train models on a shared lab GPU server reached by SSH
  • Engineers using rented Vast.ai or Modal GPUs who want a clear launch and cleanup record
  • Teams that already keep a project AGENTS.md describing their compute environment

Skip it if

  • People without access to a GPU machine: the skill only automates launching on machines you already control, and it will ask you to configure them otherwise
  • Anyone who wants an unattended run on paid cloud GPUs without review: launching and auto-destroying instances spends money and can lose data if results are not copied first
  • Non-ML tasks: the instructions are written specifically for training and experiment jobs

Try asking

  • “Run my experiment on the lab GPU server with GPU 2 and log to wandb”
  • “Deploy train.py to the Vast.ai instance and destroy it when it finishes”
  • “Check which GPUs are free on my server and start the ablation run”
  • “Launch the Modal training app with 80GB GPU and record where results are stored”
Step-by-step guide: how to use Run ML Experiments on GPU ServersInstall, first run, examples, tips and FAQs

Install in 3 steps

  1. 1

    Download the skill

    The ZIP contains the 1 original files, the MIT licence and an attribution note.

    run-experiment.zip
  2. 2

    Add it to your agent

    Claude Code (all projects)

    mkdir -p ~/.claude/skills && unzip run-experiment.zip -d ~/.claude/skills/

    Claude Code / Agent SDK (this project only)

    mkdir -p .claude/skills && unzip run-experiment.zip -d .claude/skills/

    Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.

  3. 3

    Ask for the task

    No special command needed — the agent reads the skill's description and loads it when your request matches. Try one of the prompts above.

Get better results

  • Fill in AGENTS.md with the SSH alias, conda environment, code directory, and any gpu: vast or gpu: modal settings before asking for a run, so the agent does not have to guess.
  • Review the commands the agent proposes before it runs them, especially ssh commands, rsync, wandb login, and any auto_destroy step that deletes a cloud instance.
  • Keep large data and checkpoints out of the sync path and confirm they are persisted before enabling auto-destroy.

What's inside

  • SKILL.md

Redistributed unmodified from wanshuiyin/Auto-claude-code-research-in-sleep under the MIT licence. The assessment was drafted with AI from the skill's files on 2026-10-29. Skills can instruct your agent to run code — read SKILL.md and any scripts before installing.