dataset-search

name	dataset-search
description	Search HuggingFace Hub for datasets by keyword. Use when the user wants to find training data, benchmarks, or evaluation datasets for ML/NLP/CV research.
always	false

Dataset Search

Search for datasets on HuggingFace Hub using web_fetch.

API Endpoint

GET https://huggingface.co/api/datasets

Parameters

Parameter	Type	Default	Description
search	string	required	Search keywords
limit	int	10	Number of results (1–100)
sort	string	downloads	Sort by: downloads, likes, lastModified, createdAt
direction	int	-1	-1 = descending (most popular first)
full	bool	false	Include full dataset card info

How to Use

Step 1 — Build the URL

https://huggingface.co/api/datasets?search=<query>&limit=10&sort=downloads&direction=-1

Step 2 — Call web_fetch

web_fetch("https://huggingface.co/api/datasets?search=sentiment+analysis&limit=10&sort=downloads&direction=-1")

Step 3 — Parse the JSON response

The response is a JSON array:

[
  {
    "id": "stanfordnlp/sst2",
    "downloads": 5000000,
    "likes": 320,
    "tags": ["task_categories:text-classification", "language:en"],
    "description": "The Stanford Sentiment Treebank...",
    "lastModified": "2023-10-01T00:00:00.000Z",
    "cardData": {
      "license": "mit",
      "task_categories": ["text-classification"],
      "language": ["en"]
    }
  }
]

Step 4 — Present results

For each dataset, show:

• Dataset ID (e.g. stanfordnlp/sst2) — use with HuggingFace datasets library
• Downloads — popularity indicator
• Likes
• Tags — task type, language, modality
• Description snippet
• HuggingFace page: https://huggingface.co/datasets/<id>

Examples

Search for NLP sentiment datasets

web_fetch("https://huggingface.co/api/datasets?search=sentiment+analysis&limit=10&sort=downloads&direction=-1")

Search for image classification datasets

web_fetch("https://huggingface.co/api/datasets?search=image+classification&limit=10&sort=downloads&direction=-1")

Search for question answering benchmarks

web_fetch("https://huggingface.co/api/datasets?search=question+answering&limit=10&sort=likes&direction=-1")

Get dataset details (full card)

web_fetch("https://huggingface.co/api/datasets/stanfordnlp/sst2")

How to Use a Dataset (in code)

Once you have the dataset ID, the user can load it with:

from datasets import load_dataset
ds = load_dataset("stanfordnlp/sst2")

Or download via CLI:

huggingface-cli download stanfordnlp/sst2 --repo-type dataset

Notes

• Results sorted by downloads by default — gives most widely-used datasets first
• sort=likes gives community-favorite datasets (newer, high quality)
• For private/gated datasets, user needs to authenticate with HuggingFace token
• Dataset IDs can also be used with dataset_download tool to generate a full download script