Solutions

LM-C15-C-001

A 9-token sequence needs 9 recurrent steps.

LM-C15-C-002

The hidden state has 128 numbers.

LM-C15-C-003

Five earlier positions are included.

LM-C15-C-004

The largest weight is 0.7; multiplying by 10 gives 7.

LM-C15-C-005

Yes. In standard softmax attention, the weights for one attention distribution sum to 1.