Causal LM Shift

Problem

Build the shifted input/label pair for causal language modeling: inputs are all tokens except the last and labels are all tokens except the first, so position t predicts token t + 1.

The signature is causal_lm_shift(tokens). Return [inputs, labels].

Hint

A one-token shift turns next-token prediction into ordinary token-level classification.

Test cases

case 1

in: [[1,2,3,4]]

expected: [[1,2,3],[2,3,4]]

case 2

in: [[5]]

expected: [[],[]]

case 3

in: [[1,2]]

expected: [[1],[2]]

case 4

in: [[0,0]]

expected: [[0],[0]]

Starter codePython

def causal_lm_shift(tokens):
    # Your code here
    pass

4 test cases · run your code against them in the editor