List Parquet files
Datasets can be published in any format (CSV, JSONL, directories of images, etc.) to the Hub, and they are easily accessed with the 🤗 Datasets library. For a more performant experience (especially when it comes to large datasets), the dataset viewer automatically converts every dataset to the Parquet format.
What is Parquet?
Parquet is a columnar storage format optimized for querying and processing large datasets. Parquet is a popular choice for big data processing and analytics and is widely used for data processing and machine learning.
In Parquet, data is divided into chunks called “row groups”, and within each row group, it is stored in columns rather than rows. Each row group column is compressed separately using the best compression algorithm depending on the data, and contains metadata and statistics (min/max value, number of NULL values) about the data it contains.
This structure allows for efficient data reading and querying:
- only the necessary columns are read from disk (projection pushdown); no need to read the entire file. This reduces the memory requirement for working with Parquet data.
- entire row groups are skipped if the statistics stored in its metadata do not match the data of interest (automatic filtering)
- the data is compressed, which reduces the amount of data that needs to be stored and transferred.
A Parquet file contains a single table. If a dataset has multiple tables (e.g. multiple splits or subsets), each table is stored in a separate Parquet file.
Conversion to Parquet
The Parquet files are published to the Hub on a specific refs/convert/parquet branch (like this fancyzhx/amazon_polarity branch for example) that parallels the main branch.
In order for the dataset viewer to generate a Parquet version of a dataset, the dataset must be public, or owned by a PRO user or an Enterprise Hub organization.
Using the dataset viewer API
This guide shows you how to use the dataset viewer’s /parquet endpoint to retrieve a list of a dataset’s files converted to Parquet. 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=ibm/duorc"
def query():
response = requests.get(API_URL, headers=headers)
return response.json()
data = query()The endpoint response is a JSON containing a list of the dataset’s files in the Parquet format. For example, the ibm/duorc dataset has six Parquet files, which corresponds to the test, train and validation splits of its two subsets, ParaphraseRC and SelfRC (see the List splits and subsets guide for more details about splits and subsets).
The endpoint also gives the filename and size of each file:
{
"parquet_files":[
{
"dataset":"ibm/duorc",
"config":"ParaphraseRC",
"split":"test",
"url":"https://huggingface.co/datasets/ibm/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/test/0000.parquet",
"filename":"0000.parquet",
"size":6136591
},
{
"dataset":"ibm/duorc",
"config":"ParaphraseRC",
"split":"train",
"url":"https://huggingface.co/datasets/ibm/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/train/0000.parquet",
"filename":"0000.parquet",
"size":26005668
},
{
"dataset":"ibm/duorc",
"config":"ParaphraseRC",
"split":"validation",
"url":"https://huggingface.co/datasets/ibm/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/validation/0000.parquet",
"filename":"0000.parquet",
"size":5566868
},
{
"dataset":"ibm/duorc",
"config":"SelfRC",
"split":"test",
"url":"https://huggingface.co/datasets/ibm/duorc/resolve/refs%2Fconvert%2Fparquet/SelfRC/test/0000.parquet",
"filename":"0000.parquet",
"size":3035736
},
{
"dataset":"ibm/duorc",
"config":"SelfRC",
"split":"train",
"url":"https://huggingface.co/datasets/ibm/duorc/resolve/refs%2Fconvert%2Fparquet/SelfRC/train/0000.parquet",
"filename":"0000.parquet",
"size":14851720
},
{
"dataset":"ibm/duorc",
"config":"SelfRC",
"split":"validation",
"url":"https://huggingface.co/datasets/ibm/duorc/resolve/refs%2Fconvert%2Fparquet/SelfRC/validation/0000.parquet",
"filename":"0000.parquet",
"size":3114390
}
],
"pending":[
],
"failed":[
],
"partial":false
}Sharded Parquet files
Big datasets are partitioned into Parquet files (shards) of about 500MB each. The filename contains the name of the dataset, the split, the shard index, and the total number of shards (dataset-name-train-0000-of-0004.parquet). For a given split, the elements in the list are sorted by their shard index, in ascending order. For example, the train split of the fancyzhx/amazon_polarity dataset is partitioned into 4 shards:
{
"parquet_files":[
{
"dataset":"fancyzhx/amazon_polarity",
"config":"amazon_polarity",
"split":"test",
"url":"https://huggingface.co/datasets/fancyzhx/amazon_polarity/resolve/refs%2Fconvert%2Fparquet/amazon_polarity/test/0000.parquet",
"filename":"0000.parquet",
"size":117422360
},
{
"dataset":"fancyzhx/amazon_polarity",
"config":"amazon_polarity",
"split":"train",
"url":"https://huggingface.co/datasets/fancyzhx/amazon_polarity/resolve/refs%2Fconvert%2Fparquet/amazon_polarity/train/0000.parquet",
"filename":"0000.parquet",
"size":259761770
},
{
"dataset":"fancyzhx/amazon_polarity",
"config":"amazon_polarity",
"split":"train",
"url":"https://huggingface.co/datasets/fancyzhx/amazon_polarity/resolve/refs%2Fconvert%2Fparquet/amazon_polarity/train/0001.parquet",
"filename":"0001.parquet",
"size":258363554
},
{
"dataset":"fancyzhx/amazon_polarity",
"config":"amazon_polarity",
"split":"train",
"url":"https://huggingface.co/datasets/fancyzhx/amazon_polarity/resolve/refs%2Fconvert%2Fparquet/amazon_polarity/train/0002.parquet",
"filename":"0002.parquet",
"size":255471883
},
{
"dataset":"fancyzhx/amazon_polarity",
"config":"amazon_polarity",
"split":"train",
"url":"https://huggingface.co/datasets/fancyzhx/amazon_polarity/resolve/refs%2Fconvert%2Fparquet/amazon_polarity/train/0003.parquet",
"filename":"0003.parquet",
"size":254410930
}
],
"pending":[
],
"failed":[
],
"partial":false
}To read and query the Parquet files, take a look at the Query datasets from the dataset viewer API guide.
Partially converted datasets
The Parquet version can be partial in two cases:
- if the dataset is already in Parquet format but it contains row groups bigger than the recommended size (100-300MB uncompressed). This size is better for memory usage since Parquet is streamed row group per row group in most data libraries.
- if the dataset is not already in Parquet format or if it is bigger than 5GB.
In that case the Parquet files are generated up to 5GB and placed in a split directory prefixed with “partial”, e.g. “partial-train” instead of “train”.
You can check the row groups size directly on Hugging Face using the Parquet metadata sidebar, for example here:
Parquet-native datasets
When the dataset is already in Parquet format, the data are not converted and the files in refs/convert/parquet are links to the original files. This rule suffers an exception to ensure the dataset viewer API to stay fast: if the row group size of the original Parquet files is too big, new Parquet files are generated.
Using the Hugging Face Hub API
For convenience, you can directly use the Hugging Face Hub /api/parquet endpoint which returns the list of Parquet URLs:
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://huggingface.co/api/datasets/ibm/duorc/parquet"
def query():
response = requests.get(API_URL, headers=headers)
return response.json()
urls = query()The endpoint response is a JSON containing a list of the dataset’s files URLs in the Parquet format for each split and subset. For example, the ibm/duorc dataset has one Parquet file for the train split of the “ParaphraseRC” subset (see the List splits and subsets guide for more details about splits and subsets).
{
"ParaphraseRC":{
"test":[
"https://huggingface.co/api/datasets/ibm/duorc/parquet/ParaphraseRC/test/0.parquet"
],
"train":[
"https://huggingface.co/api/datasets/ibm/duorc/parquet/ParaphraseRC/train/0.parquet"
],
"validation":[
"https://huggingface.co/api/datasets/ibm/duorc/parquet/ParaphraseRC/validation/0.parquet"
]
},
"SelfRC":{
"test":[
"https://huggingface.co/api/datasets/ibm/duorc/parquet/SelfRC/test/0.parquet"
],
"train":[
"https://huggingface.co/api/datasets/ibm/duorc/parquet/SelfRC/train/0.parquet"
],
"validation":[
"https://huggingface.co/api/datasets/ibm/duorc/parquet/SelfRC/validation/0.parquet"
]
}
}Optionally you can specify which subset name to return, as well as which split:
import requests
headers = {"Authorization": f"Bearer {API_TOKEN}"}
API_URL = "https://huggingface.co/api/datasets/ibm/duorc/parquet/ParaphraseRC/train"
def query():
response = requests.get(API_URL, headers=headers)
return response.json()
urls = query()[
"https://huggingface.co/api/datasets/ibm/duorc/parquet/ParaphraseRC/train/0.parquet"
]Each parquet file can also be accessed using its shard index: https://huggingface.co/api/datasets/ibm/duorc/parquet/ParaphraseRC/train/0.parquet redirects to https://huggingface.co/datasets/ibm/duorc/resolve/refs%2Fconvert%2Fparquet/ParaphraseRC/train/0000.parquet for example.