xlnstorch.benchmark.Benchmark

class xlnstorch.benchmark.Benchmark

Base-class that users should inherit from in order to create benchmarks.

Only 3 hooks are strictly required to be implemented:

1. device class attribute (defaults to “cpu”)

2. make_inputs()

3. forward()

Optional hooks allow for custom book-keeping around each iteration / epoch as well as post-processing of the forward result.

__init__()

Methods

__init__()
after_epoch(idx)	Symmetric counterpart of before_epoch().
before_epoch(idx)	Called before the idx-th epoch.
forward(*args, **kwargs)	The workload under test (e.g. model(*args, **kwargs)).
make_inputs()	Produce input tensors for one iteration of the benchmark.
post_forward(output)	Optional light-weight processing of the forward result.

Attributes

device