Exercises

LM-C10-C-001

A vocabulary has 6 tokens. How many logits are produced for one prediction position?

Compute it first, then check your number.

LM-C10-C-002

If a batch has 12 prediction positions, how many target tokens are used?

Compute it first, then check your number.

LM-C10-C-003

If softmax is applied over 4 vocabulary items, how many probabilities are returned?

Compute it first, then check your number.

LM-C10-C-004

If token id 9 is looked up before the model computation, how many embedding rows are selected?

Compute it first, then check your number.

LM-C10-C-005

Does cross-entropy training update model parameters through gradients? Answer 1 for yes or 0 for no.

Compute it first, then check your number.