L’interface est localisée ; les présentations et documents des Skills sont traduits automatiquement par IA dans votre langue. Le code, les commandes, chemins et termes techniques restent en version originale, et le document de l’auteur fait foi.
Explorer les Skills
Productivité · SKILL.md

experiment-launchers

experiment-launchers

Routes safe command-generation and launcher preset workflows for CUT, FastCUT, SinCUT, and pretrained experiments.

Independent GitHub snapshot · Automated classification · Original documentation governs

Documentation de la Skill

GitHub

name: experiment-launchers description: "Routes safe command-generation and launcher preset workflows for CUT, FastCUT, SinCUT, and pretrained experiments." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

experiment-launchers

Use this sub-skill when the user asks about python -m experiments, launcher presets, or how to list the concrete training/testing command strings for the repository's known experiment families.

Read this when

  • The user wants the preset commands for grumpifycat, pretrained, or singleimage.
  • The user wants to inspect commands without starting tmux panes or running training.
  • The user asks what the launcher id argument means.
  • The user hits the broken dry path in this checkout.
  • The user wants to map launcher output back to train.py or test.py commands.

What this sub-skill covers

  • Launcher CLI shape: python -m experiments <name> <cmd> <id...>.
  • Supported families: grumpifycat, pretrained, and singleimage.
  • Supported actions such as run, train, test, run_test, print_names, and print_test_names.
  • GPU selection and epoch continuation behavior in the launcher wrapper.
  • A safe bundled command-list helper that does not call tmux or execute training.

What this sub-skill does not cover

  • Actual training or testing execution. Read ../translation-workflows/ for that.
  • Dataset conversion or layout preparation. Read ../data-preparation/ for that.
  • placeholder_launcher.py as a supported workflow; it references stale or unsupported options.

Use the bundled helper

Run scripts/list_experiment_commands.py to inspect known presets safely:

python scripts/list_experiment_commands.py --family grumpifycat --kind train
python scripts/list_experiment_commands.py --family pretrained --kind test --ids 2,3
python scripts/list_experiment_commands.py --family singleimage --kind both --json

The helper prints command strings only; it never starts tmux, allocates GPUs, downloads datasets, or runs model training.

Read next

  • references/launcher-reference.md for preset command tables and CLI behavior.
  • references/troubleshooting.md for the broken dry command, missing id, and tmux/GPUtil issues.
  • ../translation-workflows/references/cli-reference.md when you need to edit the commands the launchers emit.

Quick reminders

  • print_names and print_test_names were verified as safe inspection commands.
  • The upstream dry command fails in this checkout because its method call omits required ids.
  • launch, launch_test, stop, and close can create or kill tmux windows; do not use them merely to inspect command strings.
Skills enregistrées