Lightning Talk: Implementing Single-Dim Strategies With Sharding Validator - Anshul Sinha, Meta
About this talk
This talk, presented by Anshul Sinha from Meta, explores the implementation of Single-Dim Strategies using sharding validators to address the challenges of DTensor sharding propagation. The speaker explains how the complexity of current rules combines semantic correctness with search-space pruning, which complicates the addition and fixing of op strategies. The session introduces a prototype that allows operators to specify valid placement combinations for a single mesh dimension, enabling infrastructure to efficiently expand these strategies across the full mesh. Additionally, the speaker presents a Truth Table-style sharding validator that systematically tests shapes and sharding specifications to enhance correctness and minimize unnecessary redistributions. The aim of this presentation is to facilitate faster contributions while improving overall DTensor operator coverage.
More from this event
See all 103 talks →
What PyTorch Conference Europe 2026 Was Really Like – Official PyTorchCon EU Highlights | Paris
0:53
Lightning Talk: How DeepInverse Is Solving Imaging in Science and H... Andrew Wang & Minh Hai Nguyen
9:50
Why WideEP Inference Needs Data-Parallel-Aware Scheduling - Maroon Ayoub & Tyler Michael Smith
25:37
Write Once, Run Everywhere with Pytorch Transformers - Pedro Cuenca, Hugging Face
19:17