Every number here comes from running the article's code: 5 training samples, batch_size=2, shuffle=True, drop_last=True, lr=0.5, 3 epochs, seed 123. Keys: ← → step, space plays.
5 samples ÷ batch 2 = 2 steps per epoch, 1 sample dropped
sample in current batch
queued, used or dropped
model weights
gradients
loss