GNetfc网络运行出现"段错误"

had_in

1. ==> Preparing data..
   
2. Training and validation data exist already
   
3. ==> Building model from scratch..
   
4. 
   
5. Epoch: 0
   
6. [================================================================>]  Step: 1s585ms | To 114/114 ms | Loss: 0.643 | Acc: 66.484% (605/910)
   
7. [================================================================>]  Step: 18ms | T 115/115 ms | Loss: 0.561 | Acc: 77.391% (178/230)
   
8. Saving..
   
9. * Saved checkpoint to ./checkpoint/gti_2801_gnetfc.t7
   
10. 
    
11. Epoch: 1
    
12. [================================================================>]  Step: 117ms | T 114/114 ms | Loss: 0.543 | Acc: 80.220% (730/910))
    
13. [================================================================>]  Step: 19ms | T 115/115 ms | Loss: 0.586 | Acc: 73.478% (169/230)
    
14. ==> Slicing network..
    
15. ==> Absorbing batch normalization..
    
16. ==> Merging gain..
    
17. ==> Dumping CNN parameters into .txt files..
    
18.     GTI Model gnetfc: output channel of layer 16 extended from 2 to 256
    
19. ==> Start to generate the chip file..
    
20. - 1559729304260 -140495802324736- 2-:GTI SDK 3.5.0  built on: Sep  7 2018
    
21. - 1559729304263 -140495802324736- 0-:Could not find any available device
    
22. ==> Start to perform batch inference..
    
23. 段错误 (核心已转储)

复制代码

GNet1和GNet18能够正常训练并且推断，GNetfc训练正常，推断出现上述问题，最后定位到调用gClassify出现的问题，但是该函数在封装的so文件，请问这个可能什么情况。是我配置有问题吗？训练情况下需要插入神经棒吗？另外GTI_CHIP_MODE参数是指什么，GNetfc网络的时候，GTI_CHIP_MODE参数为2，其他情况下指为0，我并没有在文档中看到相关的介绍.

rany

PLAI刚解压出来的时候就是gnetfc的配置，此时运行也会出现错误吗？
GTI_CHIP_MODE对应Image output format，可以查看nets/netConfig_2801_gnetfc.txt