is engine(input) the same as running each part of the model separately?

[tnq177]

I apologize if this question is too dumb. I'm trying to use DeepSpeed with a pretrained XLM-R model. Upon inspecting, the model consists of 2 modules: robert (the transformer layers) and lm_head (the softmax layer). Is running:

output = engine(input)

the same as:

hidden = engine.roberta(input)
output = engine.lm_head(hidden)

Thanks!