Habana® Gaudi® v0.15 Documentation¶
Before you Get Started: Be sure to read the Release Notes before installing or upgrading to 0.15.x to learn about new features and limitations.
Gaudi Architecture and Software Overview - Describes Gaudi architecture and the SynapseAI software suite.
Release Notes - Release Notes for Gaudi SynapseAI features and limitations.
Installation Guide - Describes how to obtain and install the SynapseAI software package and the TensorFlow software package.
Gaudi Migration Guide - Guides users porting their own TensorFlow or PyTorch models to the Gaudi HPU.
SynapseAI User Guides¶
TensorFlow User Guide - Provides guidelines on how to modify existing models to run on TensorFlow platform.
PyTorch User Guide - Provides guidelines on how to modify existing models to run on PyTorch platform.
Distributed Training with TensorFlow - Provides guidelines on how to run distributed training using Gaudi.
Qualification Library Guide - Provides information on the Habana Labs Qualification Tool (hl_qual) for Gaudi.
Profiler User Guide - Provides information about the SynapseAI Profiling Subsystem.
System Management Interface Tool User Guide - Describes the system management interface tool.
Debugging Guide - Provides recommendations and tips for debugging model functionality and performance.
Model Performance Optimization Guide - Provides multiple methods that can be implemented in order to achieve the best performance using the Habana Gaudi accelerator for your training models.
Kubernetes User Guide - Describes the steps to setup a generic Kubernetes solution.
TPC User Guides¶
API Reference Guides¶
Habana Communication Library (HCL) API Reference - Provides the list of Habana Communication Library APIs.
Habana Collective Communications Library (HCCL) API Reference - Provides the list of Habana Collective Communication Library APIs.
Habana Labs Management Library (HLML) API Reference - Provides the list of APIs for monitoring and managing various states within Habana Labs’ AI accelerators.
TensorFlow CustomOp API Reference - Describes APIs exposed to write custom TensorFlow Operators for the Habana Accelerator.