Embeddings Checkpoint

Test your understanding of embeddings and positional encoding.


Get every question correct to pass this checkpoint. You can review feedback, revise your answers, and submit again.

1. A table with 10 rows of width 4 receives IDs [3,3,8]. What is output shape?

Select one answer.

2. For unmasked attention without position features, permuting input rows causes what corresponding change?

Select one answer.

3. A selected row is used three times and each use contributes [1,2]. What is its lookup gradient?

Select one answer.

4. An embedding table has 7 rows and width 3. How many trainable scalars?
5. Token and position embeddings both have shape (5,4) and are added elementwise. What is the result shape?

Select one answer.

Join Discord