DSA Studio

Neural Network Abstractions 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. Linear(2,3), Tanh, Linear(3,2) form a Sequential model. How many unique trainable scalars should parameters() return?

Select one answer.

2. A child module is assigned both to self.layer and an entry of self.layers. What should the optimizer see?

Select one answer.

3. A forward pass uses Value(weight.data) instead of weight. Outputs look right but registered gradients stay zero. Why?

Select one answer.

4. What does eval() do to the linear and activation modules in this pack?

Select one answer.

Join Discord