GPU supoort in Multilabel Example?

3 次查看（过去 30 天）

显示更早的评论

Moritz Scherrmann 2020-6-19

0
链接

此问题的直接链接

https://ww2.mathworks.cn/matlabcentral/answers/551239-gpu-supoort-in-multilabel-example

编辑： Moritz Scherrmann 2020-6-19

在 MATLAB Online 中打开

Hi all,

I am currently working with the following matlab example:

https://de.mathworks.com/help/deeplearning/ug/multilabel-text-classification-using-deep-learning.html

My problem is that even though the "canUseGPU" function returns true (I am working on a Geforce RTX 2080 Ti on my local machine),

I think that the example does not run on the GPU. However, In the example, it is stated that it should run on the GPU (line 181 - 183 in the live script):

% If training on a GPU, then convert data to gpuArray.
        if (executionEnvironment == "auto" && canUseGPU) || executionEnvironment == "gpu"
            dlX = gpuArray(dlX);
        end

When i run the code and start training, the task manager tells that there is no activity on the GPU.

Furthermore, I saw that the following embedding function removes the gpu-Arrray property of the data X (line 387-397):

function Z = embedding(X, weights)
% Reshape inputs into a vector.
[N, T] = size(X, 2:3);
X = reshape(X, N*T, 1);
% Index into embedding matrix.
Z = weights(:, X);
% Reshape outputs by separating batch and sequence dimensions.
Z = reshape(Z, [], N, T);
end