uuudown / sbnn Goto Github PK

Sorry to bother, but I think there is a question in your code, according to the notes related to the kernel PackFcWeight32, the part code is to convert row-major 32-bit floating-point weight matrix into bit column-major bit-matrix, but I think it's trying to convert row-major 32-bit floating-point weight matrix into row-major bit matrixa according to your paper SBNN.

I'm so confused about that, can you check that notes one more time?

For some reason, I have only successfully run mnist_mlp. When running other models, the output is always the same. For example, when running cifar_vgg the output is

/============================
Inference_Time: 0.002 ms.
/============================
/======Label======
3 8 8 0 6 6 1 6 3 1 0 9 5 7 9 8 5 7 8 6 7 0 4 9 5 2 4 0 9 6 6 5 4 5 9 2 4 1 9 5 4 6 5 6 0 9 3 9 7 6 9 8 0 3 8 8 7 7 4 6 7 3 6 3 6 2 1 2 3 7 2 6 8 8 0 2 9 3 3 8 8 1 1 7 2 5 2 7 8 9 0 3 8 6 4 6 6 0 0 7 4 5
6 3 1 1 3 6 8 7 4 0 6 2 1 3 0 4 2 7 8 3 1 2 8 0 8 3
/=====Predict=======
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
/=====Batch Accuracy:128=======
Top-1:10.156250%
Top-5:10.156250%

I am using a RTX2080 TI, CUDA 10.0, and sm_75. The only changes to the code were setting dev to 0 and the file path of the image dataset.