training-and-experiments
Guides agents running Ludwig train, experiment, check_install, Python LudwigModel training, and multimodal or LLM training workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- training-and-experiments
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Guides agents running Ludwig train, experiment, check_install, Python LudwigModel training, and multimodal or LLM training workflows.
- disable-model-invocation
- true
02 资源文件 6
- 参考文档references/api-reference.md1.2 KB
- 参考文档references/llm-and-multimodal.md1.2 KB
- 参考文档references/troubleshooting.md1.1 KB
- 参考文档references/workflows.md1.5 KB
- 脚本scripts/check_ludwig_install.py1.6 KB
- 脚本scripts/make_tiny_tabular_project.py1.2 KB
03 SKILL.md 正文
Training and Experiments
Use this sub-skill when the task is to train or fine-tune a Ludwig model, run a full experiment, use LudwigModel.train, resume from checkpoints, understand training artifacts, or plan LLM/VLM/multimodal training.
Recommended path
- Make or validate a config and dataset first via configuration-and-data.
- Read workflows.md for CLI and Python recipes.
- Read api-reference.md when embedding Ludwig in Python.
- If the task involves LLM, VLM, LoRA/QLoRA, DPO/KTO/ORPO/GRPO, quantization, or large pretrained models, read llm-and-multimodal.md before running anything.
- For a safe local fixture:
python scripts/make_tiny_tabular_project.py --output-dir /tmp/ludwig-train-smoke
python scripts/check_ludwig_install.py --project-dir /tmp/ludwig-train-smoke --dry-run
Use --run-tiny-train only when the user approves a short training smoke.
Common commands
ludwig train --config config.yaml --dataset dataset.csv --output_directory results
ludwig experiment --config config.yaml --dataset dataset.csv --output_directory results
ludwig check_install --help
Decision points
traintrains and saves artifacts;experimenttrains and evaluates against a split.- Use
--model_resume_pathto resume if a previous run wrote checkpoints. - Use
--backend localunless Dask/Ray is explicitly needed. - Use explicit
--gpusonly after checking CUDA and memory. - Skip save/log artifacts for smoke tests; keep them for real experiments.
Route onward
- Prediction/evaluation/forecast after training: prediction-evaluation-and-inspection.
- Hyperparameter search and AutoML: automl-and-hyperopt.
- Serving/export/upload: serving-export-and-deployment.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).