Why DLD full connect layer running incorrectly?

I tried to export my vgg model to pytorch model and I finish it. Also, I trained it and save the model as a series of npy file, but when I load the file in DLD it throw some errors. The error is “ID,Origin,Source,Message3,Analysis,NvNeural,“0: Apply bias is true, but weight could not be found”” and “ID,Origin,Source,Message 4,Analysis,NvNeural,Inference failed for 4 layers”. The error is on the full connnect layer and the convolution layers have no problems. So can anyone tell me how to fix this problem? Thanks.

P.S. I do have the bias npy file in my weights folder. Futhermore,the sample in DLD also has some questions,such as the Resnet model, it does not have the bias file in the weights folder and still has bias selection been selected, so DLD gives the warning:ID,Origin,Source,Message2,Analysis,NvNeural,“RN32_RB_F16_S1_0_convolution_0: Unable to load bias weights data, random weights will be used”.

Sorry for the reply as we having issues with the notification. The error you were seeing indicates the bias data was not loaded successfully. Many factors could lead to the failure, but the first thing to check is if the bias data file is in the right format that DL Designer always expects: {layername}_bias.npy. If that is not the problem, would you mind send more details (like your bias data file and model file) to nsight-dl-feedback@nvidia.com to help us debug the problem?

Thank you for reporting this issue, we will have it fixed in the next release.

Hi, Thanks for your reply and I have sent my model file(.xml) to your e-mail address.