hard20 pts · first answer countsgpu-computingdebugging

Device-side assert in cross-entropy

Training fails with "device-side assert triggered" in nll_loss. The model has 10 output classes. What is the most likely data bug?

A.The images are not normalisedB.The learning rate is too highC.The batch size is not a power of twoD.A label value is outside 0–9, such as 10 or −1

Discussion 0

Join free to post in the discussion.