Intel Gaudi Profiler User Guide

If you are using PyTorch and are familiar with TensorBoard, you can easily start using the Intel® Gaudi® Profiler for both model development and profiling by only substituting HPU for GPU in your sources.

For a brief overview of the basic steps for PyTorch, refer to Profiling with PyTorch.

If you are interested in low-level focused profiling or writing models in Intel Gaudi software language without a framework, refer to Profiling.

For details on the profiling control flow, refer to Profiling Architecture. For further training acceleration, refer to Profiling Tips and Tricks.

• Profiling with PyTorch
• Profiling with Intel Gaudi Software
• Profiling Architecture
• Tips and Tricks to Accelerate the Training

Note

Unless otherwise specified, the terms step and batch are used interchangeably throughout this chapter.