[REQUEST] How to use `Flops Profiler` to test `model.generate()`

Is your feature request related to a problem? Please describe.

Nowadays, the Flops Profiler can test forward for transformers such as bert, but for t5, it actually uses model.generate() during inference. So how to use Flops Profiler to test model.generate() ?

Now I use this code

tokenizer = transformers.T5Tokenizer.from_pretrained('t5-base', return_dict=False)model = transformers.T5ForConditionalGeneration.from_pretrained('t5-base')

flops, macs, params = get_model_profile(
          model,
          kwargs=t5_input_constructor(batch_size, seq_len, tokenizer),
          print_profile=True,
          detailed=True,
          warm_up=10,
          module_depth=-1,
          top_modules=1,
      )

Describe the solution you'd like
A clear and concise description of what you want to happen.
To test model.generate() time cost

Describe alternatives you've considered
A clear and concise description of any alternative solutions or features you've considered.

Additional context
Add any other context or screenshots about the feature request here.