customize-data-model
Customize Lightning-Hydra-Template DataModules, LightningModules, network components, optimizers, schedulers, and Hydra target wiring.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- customize-data-model
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Customize Lightning-Hydra-Template DataModules, LightningModules, network components, optimizers, schedulers, and Hydra target wiring.
- disable-model-invocation
- true
02 资源文件 4
- 参考文档references/customization-recipes.md2.8 KB
- 参考文档references/data-model-api.md2.8 KB
- 参考文档references/troubleshooting.md2.0 KB
- 脚本scripts/check_hydra_targets.py2.7 KB
03 SKILL.md 正文
Customize Data and Model Code
Use this sub-skill when adapting the template away from the MNIST example or repairing target imports after changing package/module names.
Triggers
Read this sub-skill for tasks about:
- Replacing
MNISTDataModule, adding a newLightningDataModule, or changing data splits/loaders. - Replacing
MNISTLitModule, adding model components, changing optimizers/schedulers, metrics, ortorch.compilebehavior. - Editing
configs/data/*.yamlorconfigs/model/*.yaml_target_strings. - Renaming the default
srcpackage and fixing stale imports. - Diagnosing batch-size/world-size, metric key, dataloader dtype/shape, or target import errors.
Quick workflow
- Before editing, render the composed config and note current data/model targets.
- Add or update Python classes in the target checkout's package.
- Update config
_target_strings and constructor parameters together. - Run the bundled target checker:
python <this-skill>/sub-skills/customize-data-model/scripts/check_hydra_targets.py --repo-root . --config configs/data/mnist.yaml --config configs/model/mnist.yaml - Compose and instantiate train/eval configs without data download:
python <this-skill>/scripts/check_lightning_hydra_project.py --repo-root . --config-name train.yaml --instantiate - If the task needs real data behavior, adapt pytest fixtures or use cached/tiny data before running training tests.
Read references
- Data and model API: verified signatures, method contracts, metrics, optimizer/scheduler wiring, and config fields.
- Customization recipes: replacing data/model modules, renaming packages, and keeping configs/tests aligned.
- Troubleshooting: stale
_target_, batch-size divisibility, metric mismatch, dataloader/data-download issues, andtorch.compileproblems.
Bundled script
- scripts/check_hydra_targets.py: scans YAML configs for
_target_dotted paths, imports them from a target checkout, and optionally instantiates composed train/eval configs via the root inspector.
Boundaries
- For experiment YAML, sweeps, overrides, and paths, use configure-experiments.
- For running train/eval/checkpoints/loggers/accelerators, use train-evaluate.
- For package metadata, tests, CI, and rename gates, use test-maintain-template.
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).