# meta-pytorch/data

A PyTorch repo for data loading and utilities to be shared by the PyTorch domain libraries.

Repository: https://github.com/meta-pytorch/data
Canonical: https://ross.abutalabs.com/products/meta-pytorch-data
Language: Python
License: BSD-3-Clause
License Family: permissive
Last push: 2026-08-26T11:42:20+00:00

## Health v2 (maintenance only)
Score: 67/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 8, longevity 100
- inputs: {"age_days": 1939, "days_push": 7, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1263, forks 179 (observed 2026-08-28T04:04:10.394103+00:00)

## What it is
TorchData is a PyTorch library providing scalable, performant data loading utilities, including StatefulDataLoader, a drop-in replacement for torch.utils.data.DataLoader with mid-epoch checkpointing via state_dict/load_state_dict. It also offers torchdata.nodes, a library of composable iterators for building streaming data pipelines.

## Use cases
- checkpoint dataloader state mid-epoch for resumable training
- replace PyTorch DataLoader with a stateful version
- build streaming data loading pipelines with composable iterators
- track RNG states and worker buffers across training restarts
- share data loading utilities across PyTorch domain libraries

## When to choose
- you need fault-tolerant, resumable PyTorch training loops with mid-epoch checkpointing
- you want composable, streaming-style dataloading primitives in PyTorch
- you're on the PyTorch ecosystem and want the successor to torch.utils.data

## When to avoid
- you don't use PyTorch or its DataLoader
- you need a general-purpose ETL tool outside ML training
- you require a stable, long-frozen API since the project is still iterating

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, etl, streaming, data-science
- domain: machine-learning, deep-learning
- platform: python, cross-platform
- tags: pytorch, dataloader, checkpointing, stateful-dataloader, data-pipeline, torchdata, data-engineering, gpu

## Member repositories
- meta-pytorch/data (main) score 67

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:04:10.394103+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T05:04:09.035130+00:00, confidence not recorded.
  - readme: https://github.com/meta-pytorch/data (fetched 2026-08-28T04:04:10.394103+00:00, sha 7f5c96918694)
- Data as of 2026-08-30T08:39:29.467469+00:00.