菜单
Explore, fetch, and create Hugging Face datasets. Use the Dataset Viewer API (splits, rows, search, filter, parquet, size, statistics) to inspect data without downloading, and create or upload datasets to the Hub. Use for building or inspecting training data.
默认使用会先检查来源的 Prompt；你也可以切换为直接命令，或下载本地副本。
决定是否安装前，请先阅读 SKILL.md，以及 SkillsMP 当前展示的配套文件。
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/Abd0r/porcupineai --skill hf-datasets命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
Write code that does not look like AI output. Reject low-evidence and low-signal TypeScript patterns (chained assertions, unknown/object widening, Reflect, runtime typeof, module mocks) in favor of typed, boundary-checked code.
Control a Home Assistant smart home through its MCP server. Query sensor and entity states read-only, then act on devices (lights, switches, climate) via call_service with explicit user approval. Use for "turn off the living room lights", "what is the thermostat set to", "set bedroom temp", or any Home Assistant entity/device task.
Produce clear, accessible diagrams and charts (architecture, flow, sequence, ER, state, org, Gantt, timeline) as Mermaid or polished self-contained SVG. Covers picking the right diagram type, a consistent visual system, and accessibility. Use whenever you need to explain a system, a flow, data, or a process in a README, doc, benchmark report, plan, or launch graphic.
基于 SOC 职业分类
正在显示 SKILL.md
| name | hf-datasets |
| description | Explore, fetch, and create Hugging Face datasets. Use the Dataset Viewer API (splits, rows, search, filter, parquet, size, statistics) to inspect data without downloading, and create or upload datasets to the Hub. Use for building or inspecting training data. |
| stack | ml |
Two surfaces: the Dataset Viewer API (read-only inspection, no download)
and the Hub (create/upload). The hf datasets CLI and datasets Python
library sit on top.
Base URL: https://datasets-server.huggingface.co. GET, URL-encoded params.
Gated/private datasets need Authorization: Bearer <HF_TOKEN>.
| Purpose | Endpoint |
|---|---|
| Validate availability | /is-valid?dataset=<ns/repo> |
| List configs + splits | /splits?dataset=<ns/repo> |
| Preview rows | /first-rows?dataset=...&config=...&split=... |
| Paginate rows (max 100) | /rows?dataset=...&config=...&split=...&offset=0&length=100 |
| Full-text search | /search?dataset=...&split=...&query=<text> |
| Row predicates | /filter?dataset=...&split=...&where=<pred>&orderby=<col> |
| Parquet shard URLs | /parquet?dataset=<ns/repo> |
| Size totals | /size?dataset=<ns/repo> |
| Column statistics | /statistics?dataset=...&config=...&split=... |
Pagination: read num_rows_total / num_rows_per_page / partial from the
response to drive the next offset.
curl "https://datasets-server.huggingface.co/rows?dataset=stanfordnlp/imdb&config=plain_text&split=train&offset=0&length=100"
hf datasets info <id> # metadata
hf datasets parquet <id> # parquet URLs
hf datasets sql "<duckdb>" # SQL over parquet
datasets library to build
an Arrow dataset.hf upload <ns>/<dataset-name> <data-dir> --type dataset --commit-message "...".README.md) with task, columns, license, and a
datasets tag so it's discoverable.Inspecting a dataset without downloading it, or creating/uploading a dataset to the Hub (e.g. NANOG1 multimodal oncology data).
hf download for the
actual files.data/data-hygiene.Adapted from huggingface/skills (Apache-2.0).