Saving information about datasets

[dilyabareeva]

Problem

We want to be able to serialize datasets, without necessarily including the data itself. This is relevant for ToyBenchmarks.

Potential Solutions

• pass rootand transform to Benchmark -> the benchmarks creates torchvision.datasets.ImageFolder
• pickle the dataset object
• the user passes a dataset with load_state_dict, state_dictmethods
• add method to Benchmarks itself to handle the dataset state

[dilyabareeva]

Potential inspiration: https://github.com/deel-ai/influenciae/blob/main/deel/influenciae/benchmark/base_benchmark.py