AutoTrain Advanced

Use this skill for the Hugging Face AutoTrain Advanced repository: the top-level autotrain CLI, config-driven training, the FastAPI UI/API, backend runners, and bundled model utilities.

Start here for install, command discovery, and route selection. Then jump to the focused sub-skill that matches the task family.

Install and inspect

• Install the package in editable mode from the repo root: python -m pip install -e .
• Install a compatible PyTorch stack for your platform before GPU-backed workflows. The repo expects torch, torchvision, and torchaudio to be available.
• Minimal import check: python -c "import autotrain; print(autotrain.__version__)"
• Use scripts/check_install.py when you want a quick import/version check.
• Use scripts/inspect_cli.py --help to inspect the CLI without opening source files.
• Use scripts/check_backends.py when you need to see the current torch/CUDA view.

Route map

Read references/workflow-map.md for the command-family map, supported task families, and the one important exception: vlm is supported through the app/API/config paths, not as a top-level autotrain vlm command.

• sub-skills/cli-config/ — install, autotrain --help, --version, --config, setup, parser behavior, and config validation.
• sub-skills/llm-training/ — autotrain llm, LLM finetuning configs, quantization, PEFT, unsloth, and adapter workflows.
• sub-skills/text-and-tabular/ — text classification/regression, token classification, seq2seq, extractive QA, sentence-transformers, and tabular training.
• sub-skills/vision-multimodal/ — image classification/regression, object detection, and VLM workflows through the app/API/config path.
• sub-skills/app-backends/ — autotrain app, autotrain api, autotrain spacerunner, local/cloud backends, auth, jobs, and logs.
• sub-skills/model-tools/ — autotrain tools merge-llm-adapter and autotrain tools convert_to_kohya.

When to read deeper

• Read references/troubleshooting.md for install, import, backend, auth, and data-layout failures that affect more than one route.
• Read references/repo-provenance.md when you need to confirm whether this skill still matches the current repository checkout or when refreshing it later.
• Read the owning sub-skill before giving concrete commands, config fields, dataset checks, or backend-specific recovery steps.

Good first checks

• autotrain --help
• autotrain <subcommand> --help
• python -m pip check
• python -c "import autotrain; print(autotrain.__version__)"

Notes

• The repository is multi-modal and config-driven; route by task family, not only by source folder.
• The UI/API has broader task coverage than the top-level CLI for some workflows, especially VLM.
• Keep runtime links inside this generated skill tree; do not point future agents back at the source checkout.