MobileNetV2 training does not converge

[cherry-licongyi]

I use the mobilenetv2 code in Repository training on the CIFAR10 data set, but does not converge, do not know what reason, any answer would be appreciated

here is the training log

==> Preparing data..
==> Building model..

Epoch: 0
 [=========================== 391/391 ============================>]  Step: 2s840ms | Tot: 29s605ms | Loss: 2.304 | Acc: 9.868% (4934/50000)                                                  
 [=========================== 100/100 ============================>]  Step: 39ms | Tot: 3s794ms | Loss: 2.304 | Acc: 10.000% (1000/10000)                                                     
Saving..

Epoch: 1
 [=========================== 391/391 ============================>]  Step: 52ms | Tot: 26s718ms | Loss: 2.304 | Acc: 9.674% (4837/50000)                                                     
 [=========================== 100/100 ============================>]  Step: 42ms | Tot: 3s526ms | Loss: 2.304 | Acc: 10.000% (1000/10000)                                                     

Epoch: 2
 [=========================== 391/391 ============================>]  Step: 63ms | Tot: 27s702ms | Loss: 2.304 | Acc: 9.836% (4918/50000)                                                     
 [=========================== 100/100 ============================>]  Step: 36ms | Tot: 3s642ms | Loss: 2.303 | Acc: 10.000% (1000/10000)                                                     

Epoch: 3
 [=========================== 391/391 ============================>]  Step: 53ms | Tot: 26s19ms | Loss: 2.304 | Acc: 10.154% (5077/50000)                                                     
 [=========================== 100/100 ============================>]  Step: 36ms | Tot: 3s463ms | Loss: 2.304 | Acc: 10.000% (1000/10000)                                                     

Epoch: 4
 [=========================== 391/391 ============================>]  Step: 61ms | Tot: 26s405ms | Loss: 2.304 | Acc: 10.148% (5074/50000)                                                    
 [=========================== 100/100 ============================>]  Step: 36ms | Tot: 3s632ms | Loss: 2.304 | Acc: 10.000% (1000/10000)                                                     

Epoch: 5
 [=========================== 391/391 ============================>]  Step: 52ms | Tot: 26s824ms | Loss: 2.304 | Acc: 9.966% (4983/50000)                                                     
 [=========================== 100/100 ============================>]  Step: 34ms | Tot: 3s266ms | Loss: 2.305 | Acc: 10.000% (1000/10000)                                                     

Epoch: 6
 [=========================== 391/391 ============================>]  Step: 64ms | Tot: 25s992ms | Loss: 2.304 | Acc: 10.260% (5130/50000)                                                    
 [=========================== 100/100 ============================>]  Step: 33ms | Tot: 3s639ms | Loss: 2.304 | Acc: 10.000% (1000/10000)                                                     

Epoch: 7
 [=========================== 391/391 ============================>]  Step: 58ms | Tot: 26s53ms | Loss: 2.304 | Acc: 9.936% (4968/50000)                                                      
 [=========================== 100/100 ============================>]  Step: 34ms | Tot: 3s475ms | Loss: 2.304 | Acc: 10.000% (1000/10000)                                                     

Epoch: 8
^CTraceback (most recent call last): ..............................]  Step: 71ms | Tot: 6s349ms | Loss: 2.305 | Acc: 10.205% (1267/12416)  

[logan-mo]

Try a smaller learning rate?