Using PyTorch where it touches silicon: training and profiling models that define accelerator workloads, exporting graphs through torch.compile, ONNX or a vendor backend, writing custom operators and kernels, and quantization-aware work that drives hardware datatype and memory decisions.
No statement on what tools can do has been recorded for this skill yet.
The Hub knows of no registered course, programme or certification for this skill yet.
Note from the research register: Learned by porting and profiling real models on real accelerators in an AI hardware or compiler team, on top of the official tutorials and a machine learning foundation.
Register version 1.