PyTorch

Use it when

• train neural networks with eager Python execution and accelerator support
• build custom tensor and autograd workflows

What it solves

Not the fit when

• training infrastructure and experiment tracking are separate concerns

Install

Install the PyTorch wheel selected for the target CPU or accelerator from pytorch.org

Invoke

Define a `torch.nn.Module`, move tensors and model to the selected device, run the optimizer loop, and save a versioned state dictionary.

Alternatives

No reviewed alternatives recorded yet.