🤖 DELTASet x and y, then trigger backward() — I'll walk the graph in reverse and show you every gradient multiplication.
🧑🏫 PROF. TORCHbackward() starts with df/df = 1 at the output, then multiplies local derivatives backward through the chain rule.