Attention

June 2024 Status Update: Removing DataPipes and DataLoader V2

We are re-focusing the torchdata repo to be an iterative enhancement of torch.utils.data.DataLoader. We do not plan on continuing development or maintaining the [DataPipes] and [DataLoaderV2] solutions, and they will be removed from the torchdata repo. We’ll also be revisiting the DataPipes references in pytorch/pytorch. In release torchdata==0.8.0 (July 2024) they will be marked as deprecated, and in 0.9.0 (Oct 2024) they will be deleted. Existing users are advised to pin to torchdata==0.8.0 or an older version until they are able to migrate away. Subsequent releases will not include DataPipes or DataLoaderV2. Please reach out if you suggestions or comments (please use this issue for feedback)

S3FileLoader

class torchdata.datapipes.iter.S3FileLoader(source_datapipe: IterDataPipe[str], request_timeout_ms=- 1, region='', buffer_size=None, multi_part_download=None)

[DEPRECATED] Use https://github.com/awslabs/s3-connector-for-pytorch instead.

Iterable DataPipe that loads Amazon S3 files from the given S3 URLs (functional name: load_files_by_s3). S3FileLoader iterates all given S3 URLs in BytesIO format with (url, BytesIO) tuples.

Note

1. source_datapipe must contain a list of valid S3 URLs.

2. request_timeout_ms and region will overwrite settings in the configuration file or environment variables.

3. The lack of AWS proper configuration can lead empty response. For more details related to S3 IO DataPipe setup and AWS config, please see the README file.

Parameters:

• source_datapipe – a DataPipe that contains URLs to s3 files

• request_timeout_ms – timeout setting for each reqeust (3,000ms by default)

• region – region for access files (inferred from credentials by default)

• buffer_size – buffer size of each chunk to download large files progressively (128Mb by default)

• multi_part_download – flag to split each chunk into small packets and download those packets in parallel (enabled by default)

Example:

from torchdata.datapipes.iter import IterableWrapper, S3FileLoader

dp_s3_urls = IterableWrapper(['s3://bucket-name/folder/', ...]).list_files_by_s3()
# In order to make sure data are shuffled and sharded in the
# distributed environment, `shuffle`  and `sharding_filter`
# are required. For detail, please check our tutorial in:
# https://pytorch.org/data/main/tutorial.html#working-with-dataloader
sharded_s3_urls = dp_s3_urls.shuffle().sharding_filter()

dp_s3_files = S3FileLoader(sharded_s3_urls)
for url, fd in dp_s3_files: # Start loading data
    data = fd.read()

# Functional API
dp_s3_files = sharded_s3_urls.load_files_by_s3(buffer_size=256)
for url, fd in dp_s3_files:
    data = fd.read()