What decoder inputs is the trainer creating when I use it with AutoModelForSeq2SeqLM and a model that needs Decoder Inputs?

Beginners
1

I tried looking for the information in the docs and source code but I coudln’t find it.
Is it just feeding the inputs twice?