accelerate
A library that enables the same PyTorch code to be run across any distributed configuration. tldr-pages
On several platforms
Several platforms
Print environment information
accelerate envIdentifyInteractively create a configuration file
accelerate configIdentifyPrint the estimated GPU memory cost of running a Hugging Face model with different data types
accelerate estimate-memory name/modelIdentifyTest an Accelerate configuration file
accelerate test --config_file path/to/config.yamlIdentifyRun a model on CPU with Accelerate
accelerate launch path/to/script.py --cpuIdentifyRun a model on multi-GPU with Accelerate, with 2 machines
accelerate launch path/to/script.py --multi_gpu --num_machines 2Identify