Datasets can be published in any format (CSV, JSONL, directories of images, etc.) on the Hub, and people generally use the datasets
library to access the data. To make it even easier, the datasets-server automatically converts every dataset to the Parquet format and publishes the parquet files on the Hub (in a specific branch: ref/convert/parquet
).
This guide shows you how to use Datasets Server’s /parquet
endpoint to retrieve the list of a dataset’s parquet files programmatically. Feel free to also try it out with Postman, RapidAPI, or ReDoc
The /parquet
endpoint accepts the dataset name as its query parameter:
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://datasets-server.huggingface.co/parquet?dataset=duorc"
def query():
response = requests.request("GET", API_URL, headers=headers)
return response.json()
data = query()
The endpoint response is a JSON containing a list of the dataset’s parquet files. For example, the duorc dataset has six parquet files, which corresponds to the train
, validation
and test
splits of its two configurations (see the /splits guide):
{
"parquet_files": [
{
"dataset": "duorc",
"config": "ParaphraseRC",
"split": "test",
"url": "https://huggingface.co/datasets/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/duorc-test.parquet",
"filename": "duorc-test.parquet",
"size": 6136590
},
{
"dataset": "duorc",
"config": "ParaphraseRC",
"split": "train",
"url": "https://huggingface.co/datasets/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/duorc-train.parquet",
"filename": "duorc-train.parquet",
"size": 26005667
},
{
"dataset": "duorc",
"config": "ParaphraseRC",
"split": "validation",
"url": "https://huggingface.co/datasets/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/duorc-validation.parquet",
"filename": "duorc-validation.parquet",
"size": 5566867
},
{
"dataset": "duorc",
"config": "SelfRC",
"split": "test",
"url": "https://huggingface.co/datasets/duorc/resolve/refs%2Fconvert%2Fparquet/SelfRC/duorc-test.parquet",
"filename": "duorc-test.parquet",
"size": 3035735
},
{
"dataset": "duorc",
"config": "SelfRC",
"split": "train",
"url": "https://huggingface.co/datasets/duorc/resolve/refs%2Fconvert%2Fparquet/SelfRC/duorc-train.parquet",
"filename": "duorc-train.parquet",
"size": 14851719
},
{
"dataset": "duorc",
"config": "SelfRC",
"split": "validation",
"url": "https://huggingface.co/datasets/duorc/resolve/refs%2Fconvert%2Fparquet/SelfRC/duorc-validation.parquet",
"filename": "duorc-validation.parquet",
"size": 3114389
}
]
}
The dataset can then be accessed directly through the parquet files:
import pandas as pd
url = "https://huggingface.co/datasets/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/duorc-train.parquet"
pd.read_parquet(url).title.value_counts().head()
# Dracula 422
# The Three Musketeers 412
# Superman 193
# Jane Eyre 190
# The Thing 189
# Name: title, dtype: int64
The big datasets are partitioned in parquet files (shards) of about 1 GiB. The file name gives the index of the shard and the total number of shards. For example, the train
split of the alexandrainst/danish-wit
dataset is partitioned into 9 shards, from parquet-train-00000-of-00009.parquet
to parquet-train-00008-of-00009.parquet
:
{
"parquet_files": [
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "test",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-test.parquet",
"filename": "parquet-test.parquet",
"size": 48781933
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00000-of-00009.parquet",
"filename": "parquet-train-00000-of-00009.parquet",
"size": 937127291
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00001-of-00009.parquet",
"filename": "parquet-train-00001-of-00009.parquet",
"size": 925920565
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00002-of-00009.parquet",
"filename": "parquet-train-00002-of-00009.parquet",
"size": 940390661
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00003-of-00009.parquet",
"filename": "parquet-train-00003-of-00009.parquet",
"size": 934549621
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00004-of-00009.parquet",
"filename": "parquet-train-00004-of-00009.parquet",
"size": 493004154
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00005-of-00009.parquet",
"filename": "parquet-train-00005-of-00009.parquet",
"size": 942848888
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00006-of-00009.parquet",
"filename": "parquet-train-00006-of-00009.parquet",
"size": 933373843
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00007-of-00009.parquet",
"filename": "parquet-train-00007-of-00009.parquet",
"size": 936939176
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "train",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-train-00008-of-00009.parquet",
"filename": "parquet-train-00008-of-00009.parquet",
"size": 946933048
},
{
"dataset": "alexandrainst/danish-wit",
"config": "alexandrainst--danish-wit",
"split": "val",
"url": "https://huggingface.co/datasets/alexandrainst/danish-wit/resolve/refs%2Fconvert%2Fparquet/alexandrainst--danish-wit/parquet-val.parquet",
"filename": "parquet-val.parquet",
"size": 11437355
}
]
}
The shards can be concatenated:
import pandas as pd
import requests
r = requests.get("https://datasets-server.huggingface.co/parquet?dataset=alexandrainst/danish-wit")
j = r.json()
urls = [f['url'] for f in j['parquet_files'] if f['split'] == 'train']
dfs = [pd.read_parquet(url) for url in urls]
df = pd.concat(dfs)
df.mime_type.value_counts().head()
# image/jpeg 140919
# image/png 18608
# image/svg+xml 6171
# image/gif 1030
# image/webp 1
# Name: mime_type, dtype: int64