---
id: tfds-nightly
version: "4.9.9.dev202510250044"
license: Apache 2.0
license_treatment: permissive
maintenance: active
---

# tfds-nightly — tensorflow/datasets is a library of datasets ready to use with TensorFlow.

License: permissive · Maintenance: active · Downloads: 141.7K/mo

## What it is and what it does

tfds-nightly is a library that centralizes access to many public machine-learning datasets, downloading and preparing them into standardized tf.data.Dataset objects. It abstracts away the complexity of locating, downloading, and formatting datasets so that standard use cases work immediately—you can load a dataset like MNIST with a single function call and chain it directly into your training pipeline.

The library depends on 18 runtime packages including numpy, pyarrow, protobuf, and tensorflow-metadata to handle data serialization, array operations, and metadata management. It emphasizes simplicity, performance, determinism, and reproducibility: all users get the same examples in the same order, and the library follows best practices for data pipeline efficiency. This is a nightly build, so it tracks development versions of the underlying library.

Use it for:

- Load standard ML benchmarks without writing download or preprocessing code.
- Build reproducible training pipelines where dataset order and splits are consistent across runs.
- Prototype models quickly by chaining load output directly into data transformations.
- Access a curated catalog of public datasets with standardized metadata and documentation.
- Integrate dataset loading into workflows with minimal boilerplate.

## Worth the install?

AI-flagged interpretation of the facts above — verify before relying.

Provides a library of ready-to-use public datasets formatted as tf.data.Datasets for machine learning workflows, handling download, preparation, and standardized access.

Yes, if you are working with TensorFlow and need standard ML datasets. The library is actively maintained, has low install friction, and carries no known vulnerabilities. However, this is a nightly build (version 4.9.9.dev202510250044)—use the stable release for production unless you specifically need development features. Always verify your right to use each dataset under its own license.

## Install

pip install tfds-nightly
uv add tfds-nightly
poetry add tfds-nightly

## Installing tfds-nightly

Before you install: Low friction installation with a pure-Python wheel. Active maintenance as of 2026-07-29 with 4581 repository stars. This is a nightly build (version 4.9.9.dev202510250044), so expect development-stage stability; use the stable release for production unless you need cutting-edge features.

License in practice: Apache 2.0 permissive license allows commercial and private use with minimal restrictions. You must retain license notices and may not hold the authors liable. Note the package's own disclaimer: you are responsible for verifying your right to use each dataset under its own license.

Quickstart:

pip install tfds-nightly
import tfds_nightly
ds = tfds_nightly.load('mnist', split='train', as_supervised=True)
ds = ds.batch(128).prefetch(10)

Requires Python >=3.10. TensorFlow must be installed separately to use the loaded datasets.

Verify before relying:

- Whether this nightly build is suitable for production use or intended only for testing new dataset additions.
- Performance characteristics compared to the stable release, given the development version status.
- Complete list of available datasets in the current nightly version.

## Package facts

- License: Apache 2.0 (permissive)
- Python support: supports_current
- Install friction: low
- Maintenance: active
- Downloads: 141.7K/month (top 15,000 on PyPI)
- Known vulnerabilities: none known

## Tags

tensorflow datasets loader, ml dataset download and prepare, machine learning dataset pipeline, public datasets for tensorflow, dataset fetching and preprocessing, reproducible dataset loading, dataset catalog access, dataset-loading, ml-benchmarks

[View on SkillFed](https://skillfed.io/packages/tfds-nightly) · [View on PyPI](https://pypi.org/project/tfds-nightly/)