PyTorch Model Porting

This guide provides the initial steps and modifications needed to migrate an existing PyTorch model to Gaudi HPU. The modifications include setting up the Habana Torch Library, enabling Mixed Precision and initializing HCCL for distributed training.

This guide also provides the steps required to set up the GPU Migration toolkit in your model. The GPU Migration toolkit helps to automate the process of migrating GPU-based models to HPU, requiring minimal modifications to your model.

Using the initial steps detailed in the below sections ensures that your migrated model is functional. You can then proceed to the documents listed here for more details on the below and enable further features.

• Importing Habana Torch Library
• Enabling Mixed Precision
• Setting Up Distributed Training
• GPU Migration Toolkit

Note

This guide details the steps required to prepare PyTorch models to run on Gaudi. To create and execute a simple PyTorch model with Gaudi, see Getting Started with PyTorch and Gaudi.