Get the same loss with autograd
one move: trust automatic differentiation
Open the code pane immediately, then use the Spec tab, optional design question, tests, and artifact when you want them.
Finish each move before carrying the artifact forward.
Autograd records operations in a graph and applies the chain rule for you.
Deliverable: Write the mechanism in one sentence.Rewrite the hand-trained neuron with tensors, loss.backward(), and optimizer.step().
Deliverable: Type the smallest runnable implementation.The automated version should match the hand-derived direction and converge.
Deliverable: Record the expected output and one edge case.Inspect gradients, remove zero_grad, and compare optimizers.
Deliverable: Change one variable and explain the result.A PyTorch training script with an inspected gradient.
Deliverable: Save the code, output, and a short failure note.The automated version should match the hand-derived direction and converge.
Inspect gradients, remove zero_grad, and compare optimizers.
A PyTorch training script with an inspected gradient.
Mapped from the supplied Build Everything PDF, source page 22. The five moves are Deriva’s implementation contract for this project.