PyTorch_torchvision

EESSI_PyTorch_torchvision

Bases: RunOnlyRegressionTest, EESSI_Mixin

assert_num_ranks()

Assert that the number of reported CPUs/GPUs used is correct

pass_parallel_strategy()

Set parallelization strategy when using more than one process

set_ddp_options()

Set environment variables for PyTorch DDP

througput_per_CPU()

Training througput per device type

total_throughput()

Total training throughput, aggregated over all CPUs/GPUs