DSA Studio

Embeddings and Positional Encoding Checkpoint

Apply the module contracts to concrete traces and diagnose failures.

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

1. IDs [2,0,2] select a table, and loss sums all selected scalars. What is the per-feature gradient for table row 2?

Select one answer.

2. Token embeddings and positional embeddings both have shape (T,D). What does this pack combine_embeddings return?

Select one answer.

3. What does permutation equivariance mean for unmasked self-attention without positions?

Select one answer.

4. A lookup returns Value(table.weight[i][j].data). What important property is lost?

Select one answer.

Join Discord