training-evaluation-inference
Builds and troubleshoots PaddleDetection training, evaluation, repository inference, distributed launch, AMP, logging, and slim command workflows.
- License 许可
- Apache-2.0
- 最近更新
- 2026-10-08
- Stars
- ★ 11
- Forks
- 3
01 前言 Front Matter
- name
- training-evaluation-inference
- license
- Apache 2.0
- metadata
- { "disco-role": "operating" }
- description
- Builds and troubleshoots PaddleDetection training, evaluation, repository inference, distributed launch, AMP, logging, and slim command workflows.
- disable-model-invocation
- true
02 资源文件 3
- 参考文档references/cli-reference.md1.6 KB
- 参考文档references/training-workflows.md2.1 KB
- 参考文档references/troubleshooting.md1.5 KB
03 SKILL.md 正文
Training, Evaluation, and Repository Inference
Use this route when the task asks to train, fine-tune, resume, evaluate, run infer on images/directories, perform slice inference, use VisualDL/W&B, launch multi-GPU jobs, enable AMP/fleet, or connect config/data/weights to PaddleDetection repository tools.
Safe command planning
- Choose and summarize a config with the model-zoo/config route.
- Validate the dataset route before training or evaluation.
- Decide device and budget. CPU is fine for parser/config checks; serious training and many tutorials expect GPU.
- Construct commands with
scripts/build_train_eval_infer_command.py. It only prints commands for a user-provided checkout; it does not train. - Start with
use_gpu=falseor--device=CPUfor smoke checks, local weights, a tiny image, and--save_resultsonly when needed. - Promote to GPU/distributed/AMP only after the CPU/import/config/data preflight passes.
Common command shapes
python tools/train.py -c <config.yml> --eval -o use_gpu=true
python tools/eval.py -c <config.yml> -o weights=<model.pdparams> use_gpu=true
python tools/infer.py -c <config.yml> --infer_img=<image.jpg> --output_dir=<out> -o weights=<model.pdparams> use_gpu=false
python -m paddle.distributed.launch --gpus 0,1,2,3 tools/train.py -c <config.yml> --eval
Treat the source-checkout tools/*.py scripts as target repository entry points, not files from this generated skill. Use the bundled command builder and references for the decisions.
References
references/training-workflows.md: fine-tuning, resume, distributed launch, logging, AMP/slim, and output expectations.references/cli-reference.md: verified training/eval/infer flags and override behavior.references/troubleshooting.md: errors for devices, data, weights, configs, metrics, and optional dependencies.
04 同仓库更多 Skills
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
obsidian-clipper-template-creator
Guide for creating templates for the Obsidian Web Clipper. Use when you want to create a new clipping template, understand available variables, or format clipped content.
python-api
Programmatic Object-Detection-Metrics API guidance for BoundingBox, BoundingBoxes, Evaluator, enums, coordinate conversion, AP/mAP semantics, and source-style import diagnostics.
file-evaluation
Folder-based PASCAL VOC AP/mAP evaluation from ground-truth and detection text files.
object-detection-metrics
Guides legacy Object-Detection-Metrics PASCAL VOC AP/mAP evaluation from detection text folders and direct Python API classes.
nx-workspace-patterns
Configure and optimize Nx monorepo workspaces. Use when setting up Nx, configuring project boundaries, optimizing build caching, or implementing affected commands.