accelerate
A library that enables the same PyTorch code to be run across any distributed configuration. More information: https://huggingface.co/docs/accelerate/index.
- Print environment information:
accelerate env
- Interactively create a configuration file:
accelerate config
- Print the estimated GPU memory cost of running a Hugging Face model with different data types:
accelerate estimate-memory {{name/model}}
- Test an Accelerate configuration file:
accelerate test --config_file {{path/to/config.yaml}}
- Run a model on CPU with Accelerate:
accelerate launch {{path/to/script.py}} {{--cpu}}
- Run a model on multi-GPU with Accelerate, with 2 machines:
accelerate launch {{path/to/script.py}} --multi_gpu --num_machines 2
