The GPU Migration toolkit helps to automate the process of migrating GPU-based models to Intel Gaudi AI accelerator, requiring minimal modifications to your model. Instead of manually replacing Python API calls that have dependencies on GPU libraries, the toolkit automates this process so you can run your model with Gaudi-specific API calls, with fewer modifications.
Video
How to Migrate your models using the GPU Migration Toolkit
The GPU Migration toolkit maps specific API calls from these Python libraries and modules listed below to the appropriate equivalents in the software stack that Intel Gaudi processor will understand:
- torch.cuda
- Torch API with GPU related parameters. For example, torch.randn(device=”cuda”).
- Apex. If a specific model requires Apex, see Limitations section for further instructions.
- pynvml
The primary modification to your GPU Model involves integrating the `gpu_migration` library into your main model script. Refer to the GPU Migration Toolkit User Guide and video below for step-by-step instructions on incorporating the toolkit into your model.