Some of the problems encountered when using the GPU training network

1. OSError: [Error 12] Cannot allocate memory

Solution: This error occurs I was desperate. . . Because I read other people's solutions to add memory to resolve. . . But I am not willing, like a long - found a solution: the DataLoader is num_workersset to 1. Cons: Loading inevitably become very slow, poor my memory only 8 G

2. RuntimeError: CUDA error: device-side assert triggered

Solution: Check the category number settings are correct! Label classification should start from zero, if the total of 10 categories, the categories should be from 0 up to 19

3. CUDA error: out of memory

Solution: can be resolved appropriately reduced batch size

Guess you like

Origin www.cnblogs.com/xxxxxxxxx/p/11494734.html