Woops I ran out of memory while processing my tabular model

I have been playing with a model that barely fits on my 4 GB VM (on purpose to test something). 1 GB is already taken by other models and OS. I imported 6 identical tables that contain 2 million rows in my model using SSDT one table at the time.

I had no problem here, the model took 2.5 GB of memory. I then deployed the model to the server without processing it, just metadata. I then closed SSDT, that would unload the SSDT workspace and free up the memory. So now I have 3 GB free and a non processed database on the server.