The Agent Skills Directory

[SAFE]: The skill serves as a technical reference for PyTorch, providing code examples for building neural networks, training loops, and data pipelines. All code is standard for the PyTorch ecosystem.
[COMMAND_EXECUTION]: Documentation in SKILL.md includes the usage of the torchrun CLI tool, which is the standard utility for launching distributed training on multiple GPUs.
[DYNAMIC_EXECUTION]: The skill references torch.compile and torch.jit.script. These are core PyTorch 2.0+ features that perform runtime graph capture and compilation to optimize model performance.
[SAFE]: Example code demonstrates the use of torch.save and torch.load for managing model checkpoints. While these functions utilize pickle-based serialization, they are presented within the standard context of local model persistence.

pytorch