Recommended FL Datasets

This page lists the recommended datasets for federated learning research, which can be used with Flower Datasets flwr-datasets. To learn about the library, see the quickstart tutorial . To see the full FL example with Flower and Flower Datasets open the quickstart-pytorch.

Note

All datasets from HuggingFace Hub can be used with our library. This page presents just a set of datasets we collected that you might find useful.

For more information about any dataset, visit its page by clicking the dataset name.

Image Datasets

Image Datasets

Name	Size	Image Shape
ylecun/mnist	train 60k; test 10k	28x28
uoft-cs/cifar10	train 50k; test 10k	32x32x3
uoft-cs/cifar100	train 50k; test 10k	32x32x3
zalando-datasets/fashion_mnist	train 60k; test 10k	28x28
flwrlabs/femnist	train 814k	28x28
zh-plus/tiny-imagenet	train 100k; valid 10k	64x64x3
flwrlabs/usps	train 7.3k; test 2k	16x16
flwrlabs/pacs	train 10k	227x227
flwrlabs/cinic10	train 90k; valid 90k; test 90k	32x32x3
flwrlabs/caltech101	train 8.7k	varies
flwrlabs/office-home	train 15.6k	varies
flwrlabs/fed-isic2019	train 18.6k; test 4.7k	varies
ufldl-stanford/svhn	train 73.3k; test 26k; extra 531k	32x32x3
sasha/dog-food	train 2.1k; test 0.9k	varies
Mike0307/MNIST-M	train 59k; test 9k	32x32

Audio Datasets

Audio Datasets

Name	Size	Subset
google/speech_commands	train 64.7k	v0.01
google/speech_commands	train 105.8k	v0.02
flwrlabs/ambient-acoustic-context	train 70.3k
fixie-ai/common_voice_17_0	varies	14 versions
fixie-ai/librispeech_asr	varies	clean/other

Tabular Datasets

Tabular Datasets

Name	Size
scikit-learn/adult-census-income	train 32.6k
jlh/uci-mushrooms	train 8.1k
scikit-learn/iris	train 150

Text Datasets

Text Datasets

Name	Size	Category
sentiment140	train 1.6M; test 0.5k	Sentiment
google-research-datasets/mbpp	full 974; sanitized 427	General
openai/openai_humaneval	test 164	General
lukaemon/mmlu	varies	General
takala/financial_phrasebank	train 4.8k	Financial
pauri32/fiqa-2018	train 0.9k; validation 0.1k; test 0.2k	Financial
zeroshot/twitter-financial-news-sentiment	train 9.5k; validation 2.4k	Financial
bigbio/pubmed_qa	train 2M; validation 11k	Medical
openlifescienceai/medmcqa	train 183k; validation 4.3k; test 6.2k	Medical
bigbio/med_qa	train 10.1k; test 1.3k; validation 1.3k	Medical