Experiment Queue for GPU Batches
Runs large batches of ML experiments on remote GPU servers with retries and state tracking.
- Stars
- 18,482
- Licence
- MIT
- Files
- 3 files
What it does
When installed, the agent can take a large set of machine learning training jobs (for example, a multi-seed or multi-config grid of 10 or more) and manage them on a remote SSH GPU server. It starts jobs on free GPUs, retries jobs that run out of GPU memory after a delay, cleans up leftover terminal sessions, and launches the next group of jobs only after the previous group has finished. It also keeps a state file so progress can be checked overnight.
Great for
- ML researchers running multi-seed sweeps or hyperparameter grids on a shared SSH GPU server
- Teams that need teacher-then-student training chains to start automatically once prerequisites exist
- People who already use single-experiment tooling and now need batch orchestration for 10+ jobs
Skip it if
- Single ad-hoc experiments, where the simpler run-experiment workflow is enough
- Modal or Vast.ai deployments, which the skill says have their own orchestration
- Experiments needing manual inspection between runs, since failed jobs are parked as stuck for a human to review
Try asking
- “Run a grid of 21 seeds across 12 configurations on my GPU server and retry any out-of-memory failures”
- “Set up a teacher-then-student chain: train the teacher models, then auto-start distillation once checkpoints exist”
- “Check how many of my overnight experiment jobs are done, running, or pending”
Install in 3 steps
- 1
Download the skill
The ZIP contains the 3 original files, the MIT licence and an attribution note.
experiment-queue.zip - 2
Add it to your agent
Claude Code (all projects)
mkdir -p ~/.claude/skills && unzip experiment-queue.zip -d ~/.claude/skills/Claude Code / Agent SDK (this project only)
mkdir -p .claude/skills && unzip experiment-queue.zip -d .claude/skills/Claude apps (web/desktop): Settings → Capabilities → Skills → Upload skill, and choose the ZIP.
- 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
- Set PROJECT_DIR before running and review the generated manifest.json before launching, since the scheduler will run commands on your server.
- Do not wrap the scheduler in an extra polling loop; use its status output to check progress, as the skill itself warns about race conditions.
- Use gpu_free_threshold_mib and oom_retry settings to match a shared server's memory behavior.
What's inside
- scripts/build_manifest.py
- scripts/queue_manager.py
- 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.