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
设计与创意 · SKILL.md

generation-runtime

generation-runtime

【设计与创意】Run DataDesigner through the Python API for validate, check_models, preview, create, workflow chaining, resume, export, and runtime troubleshooting.

独立 GitHub 快照 · 自动分类 · 以原作者文档为准

技能说明

GitHub

name: generation-runtime description: "Run DataDesigner through the Python API for validate, check_models, preview, create, workflow chaining, resume, export, and runtime troubleshooting." disable-model-invocation: true metadata: disco-role: operating license: NOASSERTION

Generation Runtime

Use this sub-skill when you need to execute DataDesigner from Python rather than author column configs or map CLI commands.

Start here

  • API Reference — public DataDesigner, DatasetCreationResults, PreviewResults, RunConfig, and workflow signatures.
  • Runtime Workflows — recommended validate → readiness → preview → create flow, async usage, and workflow chaining.
  • Artifacts, Results, and Resume — artifact tree, result semantics, resume modes, and export / push-to-hub behavior.
  • Troubleshooting — model and MCP readiness failures, early shutdown, resume incompatibility, export issues, and TTY logging differences.
  • Sampler Smoke Script — deterministic local smoke that avoids remote APIs.

Use this sub-skill for

  • Running DataDesigner.validate, preview, create, acreate, check_models, set_run_config, and compose_workflow.
  • Inspecting PreviewResults and DatasetCreationResults outputs, artifact paths, processor outputs, export files, and resume checkpoints.
  • Diagnosing model alias, MCP tool alias, or async/runtime issues during local generation.
  • Understanding when validate still works even when no usable model aliases exist.

Route away

Safety defaults

  • Prefer sampler-only smoke runs first when model aliases are missing or untrusted.
  • Use validate before check_models; validate only checks the configuration shape and can succeed without usable model aliases.
  • Set RunConfig(display_tui=False, otel_metrics_port=None) when you want quiet local generation and no terminal UI.
  • Keep buffer_size, preserve_dropped_columns, and the workflow stage layout unchanged when resuming a run.
  • Do not call push_to_hub from the smoke helper; it requires network access and a valid Hugging Face token.
我的收藏