TPUs (via XLA integrations)

E96636

PyTorch accelerator backend XLA-based compilation target hardware accelerator integration

TPUs (via XLA integrations) are Google's specialized tensor processing units that can be used as accelerators for PyTorch models through the XLA compilation framework.

Jump to: Surface forms Statements Referenced by

Observed surface forms (3)

Surface form	Occurrences
PyTorch/XLA runtime	1
TPUs	1
XLA (Accelerated Linear Algebra)	1

Statements (47)

Predicate	Object
instanceOf	PyTorch accelerator backend ⓘ XLA-based compilation target ⓘ hardware accelerator integration ⓘ
abstracts	low-level TPU device management ⓘ
aimsTo	accelerate deep learning workloads ⓘ reduce training time for large models ⓘ
benefits	users needing scalable training on Google Cloud TPUs ⓘ
category	hardware-accelerated deep learning backend ⓘ machine learning infrastructure ⓘ
compatibleWith	Google Cloud TPU V2 ⓘ Google Cloud TPU V3 ⓘ Google Cloud TPU V4 ⓘ
designedFor	high-throughput tensor operations ⓘ large batch training ⓘ
developedBy	Google ⓘ
documentationHostedAt	https://github.com/pytorch/xla ⓘ
enables	accelerated tensor computations ⓘ execution of PyTorch models on TPUs ⓘ graph compilation via XLA ⓘ
exposes	XLA-specific debugging tools ⓘ profiling utilities for TPU workloads ⓘ
handles	automatic differentiation on TPU via XLA graphs ⓘ
integratesWith	PyTorch autograd system via XLA ⓘ
mapsTo	TPU cores as PyTorch devices ⓘ
optimizationMethod	ahead-of-time compilation ⓘ graph-level optimization ⓘ operation fusion ⓘ
partOf	XLA ⓘ surface form: PyTorch/XLA project ecosystem
provides	PyTorch-like APIs for TPU execution ⓘ device placement utilities ⓘ distributed data loader support ⓘ
requires	TPUs (via XLA integrations) self-linksurface differs ⓘ surface form: PyTorch/XLA runtime XLA compiler ⓘ XLA-compatible PyTorch operations ⓘ specialized input pipelines for TPUs ⓘ
supports	data parallel training ⓘ distributed training ⓘ mixed precision training ⓘ model parallel training ⓘ synchronous data parallelism across TPU cores ⓘ
supportsFramework	PyTorch ⓘ PyTorch ⓘ surface form: PyTorch/XLA
targetHardware	Tensor Processing Unit ⓘ surface form: Google TPU
usedFor	inference of deep learning models ⓘ training neural networks ⓘ
usedIn	large-scale machine learning experiments ⓘ
usesFramework	XLA ⓘ

Referenced by (4)

Full triples — surface form annotated when it differs from this entity's canonical label.

AlphaZero → hardwareUsed → TPUs (via XLA integrations) ⓘ

this entity surface form: TPUs

TPU → programmedWith → TPUs (via XLA integrations) ⓘ

this entity surface form: XLA (Accelerated Linear Algebra)

TPUs (via XLA integrations) → requires → TPUs (via XLA integrations) self-linksurface differs ⓘ

this entity surface form: PyTorch/XLA runtime

PyTorch → supportsHardware → TPUs (via XLA integrations) ⓘ