If you’re new to using Intel Gaudi AI accelerators, this page is designed for you. The Quick Start Guide will walk you through the steps to access the Intel AI Cloud and run models from the Intel Gaudi Model References and Hugging Face repositories. It includes examples for single card, multi-card, and DeepSpeed examples
- If you plan to access Intel Gaudi processors with the Intel AI Cloud, please refer to the Quick Start Guide and the accompanying video below.
- If you have access to Intel Gaudi processors via an on-premise node or another cloud provider, please follow the vendor’s instructions* for accessing the node. Then, proceed directly to the Start Training section of the Quick Start guide to launch an Intel Gaudi Docker image and start running models.
* For more information on node setup please refer to the installation guide
Basic Questions for New Customers
The performance numbers are located on the developer website here. There is a tab for both training and Inference
Users should start with the Hugging Face models from the Optimum Habana library.
Use the Tensorboard feature with profiling turned on
Use the GPU migration Toolkit
Getting Support
- Utilize Habana’s Jira Service Desk to file and communicate on issues, share test scripts and logs (for NDA customers).
- Visit the public Intel Gaudi discussion Forum for a generic queries (there is Q&A section provisioned as well).
- For Intel DevCloud Support: infrastructure related matters such as accounts, network access and storage, and environment, reach out to Intel AI Cloud Helpdesk JIRA.
Video
Onboarding with Intel Gaudi processors