training-recipes
Diffusers training recipes for DreamBooth, LoRA, textual inversion, ControlNet, text-to-image, T2I-Adapter, InstructPix2Pix, SDXL, Flux, and SD3 examples with dataset and Accelerate safety checks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- training-recipes
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Diffusers training recipes for DreamBooth, LoRA, textual inversion, ControlNet, text-to-image, T2I-Adapter, InstructPix2Pix, SDXL, Flux, and SD3 examples with dataset and Accelerate safety checks.
- disable-model-invocation
- true
02 资源文件 5
- 参考文档references/datasets-and-launch.md4.6 KB
- 参考文档references/recipes.md6.2 KB
- 参考文档references/troubleshooting.md5.2 KB
- 脚本scripts/dataset_layout_check.py7.7 KB
- 脚本scripts/training_command_builder.py11.0 KB
03 SKILL.md 正文
Diffusers Training Recipes
Use this sub-skill when the task is to prepare or adapt Diffusers training examples, build an accelerate launch command, validate a local image dataset, choose between DreamBooth/LoRA/textual inversion/ControlNet/T2I-Adapter/text-to-image recipes, or debug training setup failures.
Route elsewhere:
- Inference-only pipeline usage, loading trained adapters, or generation scripts: use
pipelines-and-inference. - Maintainer test policy, copied-code updates, or conversion PR mechanics: use
conversion-and-maintenance.
Fast Path
- Identify the recipe and base family from the user request: SD 1.x/2.x, SDXL, SD3, Flux, or a task-specific adapter.
- Validate local datasets before composing commands:
python sub-skills/training-recipes/scripts/dataset_layout_check.py --data-dir DATA --require-captions- Add
--conditioning-dir CONDfor ControlNet/T2I-Adapter/Flux control image pairs.
- Generate a dry-run command rather than launching training:
python sub-skills/training-recipes/scripts/training_command_builder.py --recipe dreambooth-lora --model MODEL --dataset DATA --output-dir OUT --instance-prompt "a photo of sks dog"
- Add memory controls before suggesting real runs:
--mixed_precision,--gradient_checkpointing, small--train_batch_size,--gradient_accumulation_steps, checkpointing, and a capped--max_train_steps. - Require explicit user confirmation before running expensive training, Hub pushes, remote downloads of gated models, or commands that overwrite a non-empty output directory.
References
- Recipe selection and command patterns: references/recipes.md
- Dataset formats and validation: references/datasets-and-launch.md
- Troubleshooting training failures: references/troubleshooting.md
Bundled Helpers
scripts/training_command_builder.pyprints safeaccelerate launchcommands for common recipes without importing Diffusers or starting training.scripts/dataset_layout_check.pyvalidates tiny local imagefolder layouts, optional captions, metadata files, and conditioning-image pair counts.
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).