chickadee » llama » transformer-weights->am-weights

Converts weights in either layout to the layout of this module: one array morphism per matrix and layer, with Wq, Wk and Wv stacked into one matrix per layer.