data-and-configs
Select, validate, and safely modify D-FINE dataset and YAML configs.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- data-and-configs
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Select, validate, and safely modify D-FINE dataset and YAML configs.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/data-and-configs.md12.7 KB
- 参考文档references/troubleshooting.md6.9 KB
- 脚本scripts/validate_detection_dataset.py23.5 KB
03 SKILL.md 正文
D-FINE data and configs router
Use this sub-skill when the task is about D-FINE dataset layout, annotation sanity, choosing a COCO/Objects365/CrowdHuman/VOC/custom config, changing YAML fields, adjusting class counts, remap policy, batch size, input size, or doing a safe dataset/config preflight.
Start with references/data-and-configs.md. For failures, use references/troubleshooting.md. For COCO-format annotation checks, run the bundled validator in scripts/validate_detection_dataset.py.
Route here for
- Selecting a model-size config for COCO, Objects365, Objects365-to-COCO, Objects365-to-custom, CrowdHuman, VOC, or custom COCO-format data.
- Explaining how
__include__,task,num_classes,remap_mscoco_category, dataloader fields, transforms,total_batch_size, and input size interact. - Preparing a custom dataset config safely before training or evaluation.
- Preflighting Object365 remap/resize assumptions without running long or mutating dataset scripts.
- Diagnosing malformed COCO JSON, missing images/annotations, wrong category IDs, class-count mistakes, or batch-size/world-size divisibility errors.
Route away
- Training, evaluation, resume, tuning, checkpoints, DDP launch, AMP, or output directories: use ../training-evaluation/SKILL.md.
- PyTorch/ONNX/OpenVINO/TensorRT inference, export, or benchmark commands that reuse a config: use ../inference-export/SKILL.md.
- Model internals, registry construction, class-head mapping code, component changes, or deploy-mode behavior: use ../architecture-api/SKILL.md.
Quick operating loop
- Identify the dataset family and desired model size (
n,s,m,l, orx). - Pick the nearest config family from the catalog in references/data-and-configs.md.
- For custom COCO-format data, set
num_classes,remap_mscoco_category, train/valimg_folder, and train/valann_filebefore constructing a training command. - Validate annotations with
python scripts/validate_detection_dataset.py --annotation <instances.json> --image-root <images-dir> --dataset-config <dataset-or-final-config.yml>. - If a workflow proceeds to launch training/evaluation or inference/export, hand off to the sibling sub-skill instead of continuing here.
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).