Quiz · 8 questions
🔄 The Training Loop
Five steps. The same five steps. Forever. Master them and the rest is engineering.
Level 0Tensor Curious
0 XP0/62 lessons0/13 achievements
0/120 XP to next level120 XP to go0% complete
Quiz
01What is the correct order of the five training-loop steps?
02What should you pass to
nn.CrossEntropyLoss as the prediction?03Which is the modern recommended optimizer for most deep-learning tasks?
04What does
pin_memory=True in DataLoader do?05Why does the Transformer convention exclude bias and norm parameters from weight decay?
Hint
Think about what L2 regularization is supposed to penalize and which params it makes sense for.
06Per-step schedulers like OneCycleLR should be stepped...
07When loading a saved model, why use
weights_only=True?08What's the 'overfit one batch' debugging trick?
Comments 0
🔔 Reply notifications (sign in)Sign in — Please sign in to comment.
No comments yet — be the first.