data-and-training
Guide dataset layout, batch preprocessing outputs, ImageFolder loader semantics, GAN training CLI/options, checkpoint/resume behavior, and validation before training.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- data-and-training
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Guide dataset layout, batch preprocessing outputs, ImageFolder loader semantics, GAN training CLI/options, checkpoint/resume behavior, and validation before training.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/data-formats.md3.5 KB
- 参考文档references/training-workflow.md5.3 KB
- 参考文档references/troubleshooting.md3.8 KB
- 脚本scripts/build_training_command.py5.4 KB
- 脚本scripts/validate_dataset_layout.py8.1 KB
03 SKILL.md 正文
Data and Training
Use this sub-skill when you need to prepare Photo2Cartoon training data, validate the dataset/photo2cartoon layout, understand preprocessing outputs, or review the GAN training launch and checkpoint flow.
Route away from this sub-skill when the main topic is:
- generator/discriminator architecture or tensor internals:
../model-internals/SKILL.md - face alignment, segmentation, or crop math:
../preprocessing/SKILL.md .pt/.onnxinference assets or prediction flow:../portrait-inference/SKILL.md
Start Here
- Read
references/data-formats.mdfor the dataset tree, supported image extensions,ImageFoldersemantics, and preprocessing output naming. - Read
references/training-workflow.mdfor thetrain.pyCLI, defaults, result-directory naming, checkpoint keys, and resume behavior. - Read
references/troubleshooting.mdwhen a split is missing, a folder is empty, an extension is ignored, a checkpoint load fails, or GPU memory is tight. - Run the bundled validator before training:
python scripts/validate_dataset_layout.py --dataset-root /path/to/dataset/photo2cartoon
- Build the guarded training command from this skill instead of copying a source-script command by hand:
python scripts/build_training_command.py --repo-root /path/to/photo2cartoon-checkout --dataset photo2cartoon
Use --strict when unsupported files should fail the dataset check and --check-images when Pillow decode validation is useful.
Covered Facts
dataset/photo2cartoon/{trainA,trainB,testA,testB}layout- allowed image suffixes and recursive
ImageFolderscanning - preprocessing output naming and white-background face crops
train.pydefaults, result-directory naming, and checkpoint keys--resume/--pretrained_weightsexpectations- loss weights, GPU limits, and batch-size cautions
Not Covered
- generator/discriminator internals
- face detection / segmentation implementation details
- inference recipes for model assets
- network downloads, training runs, or file writes outside the bundled validator
Bundled Helper
scripts/validate_dataset_layout.py: safe, argparse-based dataset layout validator with explicit--dataset-rootand optional stricter checks.scripts/build_training_command.py: guarded command builder for source-compatible train/test launches; dry-run by default and executes only with--execute.
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).