计划执行与检查节点
executing-plans
按既定实施计划完成工作,保持过程检查和反馈。
pipelines-and-models【效率与协作】Use ModelScope inference pipelines, model/preprocessor registries, output keys, and safe local custom pipeline smoke checks.
独立 GitHub 快照 · 自动分类 · 以原作者文档为准
Use this sub-skill when a task needs ModelScope inference or registry behavior:
constructing pipeline(...), loading Model.from_pretrained(...) or
Preprocessor.from_pretrained(...), choosing task constants, reading standard
OutputKeys, validating batching behavior, or creating a local custom pipeline
smoke check.
pipeline_name, device, revision, optional dependency,
and trust_remote_code choices.Tasks, input checking, output key conventions, and safe result handling.This sub-skill covers inference and registry workflows only. For template-file
creation with ModelScope CLI, route to ../customization-and-development/SKILL.md.
For training, evaluation loops, metrics-in-trainer usage, or checkpoint training
outputs, route to ../training-and-evaluation/SKILL.md. For detailed dataset
loading, MsDataset.load(...), dataset schemas, or dataset cache behavior, route
to ../datasets-config/SKILL.md.
CUDA, domain-specific accelerator execution, and heavyweight model downloads are optional and unverified in this production scope. Prefer explicit CPU execution for portable smoke checks.