Build or migrate Gaudi AI training models with these guides, videos and tools.
The Gaudi® platform architecture is purpose-designed for Deep Learning training workloads in data centers. It comprises a fully programmable Tensor Processing Core (TPC) cluster with supporting development tools and libraries. With Gaudi’s programmable architecture, we give the data scientist flexibility and ease of use to migrate or build models with Habana’s SynapseAI® software stack, reference models, extensive kernel libraries and documentation.
Here are resources to help Data Scientists bring up models to run on Gaudi. The assumption is the environment has been configured and ready for model training. If additional system configuration is required, refer to the Setup and Install page to get started.
Docs
Installation Guide
Detailed instructions for installation of the Habana Gaudi driver and SynapseAI software stack if you need to setup your environment.
TensorFlow User Guide
Detailed description of how to run your model on TensorFlow.
PyTorch User Guide
Detailed description of how to run your model on PyTorch.
Distributed Training with TensorFlow
Learn how to scale your TensorFlow models training with Horovod and HPUStrategy APIs support and Gaudi NIC or Host NIC for system scaling.
Distributed Training with PyTorch
Learn how to scale your PyTorch models training with Distributed Data Parallel APIs support and Gaudi NIC or Host NIC for system scaling.
GitHub
Habana Models
View repository of all Habana reference models and instructions on how to run and train your models, including TensorFlow and PyTorch.
Videos
Gaudi: Model Migration
A how-to tutorial to make minimum changes needed to get your model to run on Gaudi.
Gaudi: Training with TensorFlow
A how-to tutorial to help you run your model on TensorFlow.
Gaudi: Training with PyTorch
A how-to tutorial to help you run your model on PyTorch