# GetBatch: Distributed Multi-Object Retrieval for ML Data Loading

arXiv: https://arxiv.org/abs/2602.22434 · HTML: https://mlsanity.com/paper/2602.22434/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 7 of its 15 references that are in the map.
- First posted on arXiv 25 Feb 2026.

## Builds on (most cited first, 7 of 7)

- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [NeMo: a toolkit for building AI applications using Neural Modules](https://mlsanity.com/paper/1909.09577/) (2019)
- [FFCV: Accelerating Training by Removing Data Bottlenecks](https://mlsanity.com/paper/2306.12517/) (2023)
- [Analyzing and Mitigating Data Stalls in DNN Training](https://mlsanity.com/paper/2007.06775/) (2020)
- [Lhotse: a speech data representation library for the modern deep learning ecosystem](https://mlsanity.com/paper/2110.12561/) (2021)
- [High Performance I/O For Large Scale Deep Learning](https://mlsanity.com/paper/2001.01858/) (2020)
- [Training and Inference Efficiency of Encoder-Decoder Speech Models](https://mlsanity.com/paper/2503.05931/) (2025)

## Abstract (arXiv)

Machine learning training pipelines consume data in batches. A single training step may require thousands of samples drawn from shards distributed across a storage cluster. Issuing thousands of individual GET requests incurs per-request overhead that often dominates data transfer time. To solve this problem, we introduce GetBatch - a new object store API that elevates batch retrieval to a first-class storage operation, replacing independent GET operations with a single deterministic, fault-tolerant streaming execution. GetBatch achieves up to 15x throughput improvement for small objects and, in a production training workload, reduces P95 batch retrieval latency by 2x and P99 per-object tail latency by 3.7x compared to individual GET requests.