encrypted-datasets 1.0.6

Convenience functions for symetrically encrypting/decrypting huggingface Datasets

Installation

pip install encrypted-datasets

Usage

Raw string key

from datasets import load_dataset
from encrypted_datasets import encrypt_dataset, decrypt_dataset

huggingface_api_token = 'API_TOKEN'
downloaded_dataset = load_dataset('organization/dataset_repo', token=huggingface_api_token)
key = 'Your Symetric encryption key'

decrypted_dataset = decrypt_dataset(downloaded_dataset, key)

# Make modifications to decrypted_dataset...

re_encrypted_dataset = encrypt_dataset(decrypted_dataset, key)

re_encrypted_dataset.push_to_hub('organization/dataset_repo',token=huggingface_api_token)

AWS Key management service key

In this method, you use an AWS KMS key to encrypt data keys that are stored in huggingface with the data.

Create new encryped dataset and upload it to huggingface hub

from encrypted_datasets import EncryptedDataset, KMSCypher
import boto3

kms_client = boto3.client('kms')
kms_key_id = '<KMS_KEY_ID>'

cypher = KMSCypher(
    key_id=kms_key_id,
    client=client
)

dataset = Dataset.from_pandas(...)

encrypted_dataset = EncryptedDataset.encrypt(dataset, cypher)

encrypted_dataset.push_to_hub('organization/repo_id', token='<ACCESS_TOKEN>')

Load encrypted dataset, modify it, and reupload

from encrypted_datasets import EncryptedDataset, KMSCypher
import boto3

kms_client = boto3.client('kms')
kms_key_id = '<KMS_KEY_ID>'
hf_token= '<HF_TOKEN>'

cypher = KMSCypher(
    key_id=kms_key_id,
    client=client
)

encrypted_dataset = EncryptedDataset.load('organization/repo_id', token=hf_token)

dataset = encrypted_dataset.decrypt(cypher)

# Make modifications to dataset...


new_encrypted_dataset = EncryptedDataset.encrypt(dataset, cypher)

new_encrypted_dataset.push_to_hub('organization/repo_id', token=hf_token)