Distributed PyTorch training with less code.
Guides conversion of PyTorch training scripts to distributed execution with automatic device placement and mixed precision.
When to use it
Use when simplifying distributed training, sharing one script across hardware, or quickly prototyping in the HuggingFace ecosystem.
It changes how the agent writes PyTorch training scripts by using Accelerate to prepare training objects and perform backpropagation.
What you provide
Declares accelerate as a dependency and uses its Accelerator API throughout the guide.
Declares torch as a dependency for the PyTorch training scripts.
This skill
accelerate
Installs accelerate
~/.cache/huggingface/accelerate/default_config.yaml
Saves your Accelerate configuration