Implementation:Huggingface Datasets DatasetInfo

Knowledge Sources	Huggingface Datasets HF Datasets Docs
Domains	Data_Engineering, NLP
Last Updated	2026-02-14 18:00 GMT

Overview

Concrete tool for configuring dataset metadata (description, citation, license, features) provided by the HuggingFace Datasets library.

Description

DatasetInfo is a dataclass that documents datasets, including name, version, features, description, citation, license, split information, and size statistics. It supports serialization to/from JSON for persistence alongside data files. A subset of fields (config_name, download_size, dataset_size, features, splits) is included in dataset card YAML front matter. The class automatically converts raw dict/list representations to proper typed objects (Features, SplitDict, Version, etc.) in its __post_init__.

Usage

Use DatasetInfo to attach metadata to datasets during creation, or access it via dataset.info on existing datasets. It is automatically populated and serialized during push_to_hub.

Code Reference

Source Location

• Repository: datasets
• File: src/datasets/info.py
• Lines: 92-323

Signature

@dataclass
class DatasetInfo:
    description: str = dataclasses.field(default_factory=str)
    citation: str = dataclasses.field(default_factory=str)
    homepage: str = dataclasses.field(default_factory=str)
    license: str = dataclasses.field(default_factory=str)
    features: Optional[Features] = None
    post_processed: Optional[PostProcessedInfo] = None
    supervised_keys: Optional[SupervisedKeysData] = None
    builder_name: Optional[str] = None
    dataset_name: Optional[str] = None
    config_name: Optional[str] = None
    version: Optional[Union[str, Version]] = None
    splits: Optional[SplitDict] = None
    download_checksums: Optional[dict] = None
    download_size: Optional[int] = None
    post_processing_size: Optional[int] = None
    dataset_size: Optional[int] = None
    size_in_bytes: Optional[int] = None

Import

from datasets import DatasetInfo

I/O Contract

Inputs

Name	Type	Required	Description
description	str	No	A textual description of the dataset.
citation	str	No	A BibTeX citation string.
homepage	str	No	URL to the dataset's official homepage.
license	str	No	License name or terms.
features	Features	No	The dataset's column types schema.
version	str or Version	No	Version identifier for the dataset.
splits	SplitDict	No	Mapping of split names to split metadata.

Outputs

Name	Type	Description
instance	DatasetInfo	A DatasetInfo object containing all metadata fields.

Usage Examples

Basic Usage

from datasets import DatasetInfo, Features, Value

info = DatasetInfo(
    description="A sample dataset for demonstration.",
    citation="@article{sample, title={Sample}}",
    license="MIT",
    features=Features({"text": Value("string"), "label": Value("int32")}),
)

# Pass to dataset construction
from datasets import Dataset
ds = Dataset.from_dict(
    {"text": ["Hello"], "label": [1]},
    info=info,
)
print(ds.info.description)
# "A sample dataset for demonstration."

Related Pages

Implements Principle

• Principle:Huggingface_Datasets_Dataset_Metadata_Configuration