界面已本地化;技能介绍与说明文档按所选语言由 AI 自动翻译,代码、命令、路径和专业术语保留原文,以原作者文档为准。
探索技能
开发与工程 · SKILL.md

training-and-cli

training-and-cli

【开发与工程】Config-driven training, conversion, debugging, and packaging workflows for spaCy.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: training-and-cli description: "Config-driven training, conversion, debugging, and packaging workflows for spaCy." disable-model-invocation: true metadata: disco-role: operating license: MIT

training-and-cli

Enable config-driven training, evaluation, packaging, and safe CLI/data-conversion workflows for spaCy without large training runs or downloads by default.

Use this sub-skill for

  • generating and completing training configs;
  • validating config, data, and package compatibility;
  • converting IOB, CoNLL, JSON, and tiny textcat JSONL data to .spacy;
  • bounded CPU-first training and evaluation smoke checks.

Read

Route out

  • Project orchestration and spacy project commands go to project-workflows.
  • Custom component and factory Python implementation goes to pipeline-components.
  • Doc, DocBin, tokenizer, matcher, and visualization details go to documents-and-visualization.
  • Installation, import, backend, and extra-dependency issues go to install-and-inspect.

Safe default order

  1. init config or init fill-config
  2. debug config
  3. convert
  4. debug data
  5. train
  6. evaluate
  7. package
  8. validate

Triage rule

  • If debug config fails, fix config or registry references first.
  • If debug config passes but debug data fails, inspect paths, corpus files, or annotations next.

Never run full training by default; keep training smoke tests tiny and CPU-first unless the environment explicitly proves otherwise.

我的收藏