“The largest version of the Flamingo model has a total of 80 billion parameters, consisting of a frozen 70-billion-parameter Chinchilla model and an additional 10 billion trainable parameters for vision.”