PyTorch

Has run a PyTorch tutorial; understands tensors are like NumPy arrays on GPU.

Builds and trains nn.Module models with DataLoader; moves tensors between CPU/GPU; uses Optimizer and loss functions.

Writes custom datasets, custom layers, training loops with mixed precision; uses torch.compile and profiling.

Implements custom CUDA kernels via torch.autograd.Function; applies DistributedDataParallel; debugs memory and gradient issues.

Contributes to PyTorch core, custom backends, or novel training frameworks.