hest.bench.benchmark

hest.bench.benchmark(encoder: Module | None, enc_transf: Callable | None, precision: dtype | None, cli_args: dict = None, **kwargs) → Tuple[list, dict]

Benchmark a patch encoder on HEST-bench

Parameters:

• encoder (torch.nn.Module) – patch encoder to benchmark. Can be: - a plain torch module (requires enc_transf and precision), or - a TRIDENT-style encoder object exposing eval_transforms and precision.

• enc_transf (Callable) – transformation applied to encoder during inference (only required when encoder is a plain torch module).

• precision (torch.dtype) – precision used by torch.amp.autocast(‘cuda’) during inference (only required when encoder is a plain torch module).

• cli_args (dict) – cli_arguments. Defaults to None.

• **kwargs – lookup BenchmarkConfig for additional parameters