set eval and train modes before benchmarking a model (#226)
* add set_mode
* revert changes to test.py
* revert inadvertant changes
* add base class
* fix adding_models.md
* add a base model
* fix ident in pytorch_struct
* Fixes
* revert darknet
* fix demucs
* address feedback
