Causal LM Shift

Prepare next-token-prediction data from a sequence of token ids: the inputs are all tokens but the last, and the targets are all tokens but the first, so the model reads token $tt$ and must predict token $t+1t+1$.

Implement solve(ids).

Returns. Return a tuple (inputs, targets) of NumPy arrays, each one shorter than ids.

Examples

Example 1

Input:

solve([10, 11, 12, 13])

Output:

([10, 11, 12], [11, 12, 13])

Example 2

Input:

solve([7, 8])

Output:

([7], [8])