Am I wrong? demo.pyinb Input[7] didn't update total_loss.data after the last learning.

# optimization
for k in range(100):

# forward
total_loss, acc = loss()

# backward
model.zero_grad()
total_loss.backward()

# update (sgd)
learning_rate = 1.0 - 0.9*k/100
for p in model.parameters():
p.data -= learning_rate * p.grad

if k % 1 == 0:
# should it add a new line here?
# total_loss, acc = loss()
print(f"step {k} loss {total_loss.data}, accuracy {acc*100}%")

Open demo.pyinb at Input[7] and trace when total_loss is computed relative to the parameter update in the loop. Compare the printed total_loss with a fresh loss evaluation after the update; done means the notebook’s behavior and expected output are clarified.

Written by the indexing model from the issue text.