Cannot use GPU when output.shape[1] * nnz(a) > 2^31

[ahukui]

When I use the gcn layer, I always meets this problem.

InvalidArgumentError: 2 root error(s) found.
(0) Invalid argument: Cannot use GPU when output.shape[1] * nnz(a) > 2^31
[[node graph_convolution_23/SparseTensorDenseMatMul/SparseTensorDenseMatMul (defined at /home/labadmin/anaconda2/lib/python2.7/site-packages/keras/backend/tensorflow_backend.py:1083) ]]
[[loss_3/add_2/_677]]
(1) Invalid argument: Cannot use GPU when output.shape[1] * nnz(a) > 2^31
[[node graph_convolution_23/SparseTensorDenseMatMul/SparseTensorDenseMatMul (defined at /home/labadmin/anaconda2/lib/python2.7/site-packages/keras/backend/tensorflow_backend.py:1083) ]]
0 successful operations.
0 derived errors ignored.

[zhouchunpong]

I meet this problem too ！

[cquzys]

hello, do you solve it? can you share your method? thanks

[ahukui]

No

[ahukui]

You can run your code on CPU.

…

------------------&nbsp;原始邮件&nbsp;------------------ 发件人:&nbsp;"Horace"<[email protected]&gt;; 发送时间:&nbsp;2019年12月14日(星期六) 中午11:06 收件人:&nbsp;"tkipf/keras-gcn"<[email protected]&gt;; 抄送:&nbsp;"kui"<[email protected]&gt;;"Author"<[email protected]&gt;; 主题:&nbsp;Re: [tkipf/keras-gcn] Cannot use GPU when output.shape[1] * nnz(a) &gt; 2^31 (#51) hello, do you solve it? can you share your method? thanks — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe.

[innekemayachita]

Is there any solution on this issue?

[icoric4]

@innekemayachita Have you found anything maybe?

[YingtongDou]

You can cast the sparse matrix to dense matrix and use tf.matmul

[b3326023]

Same problem here.

[b3326023]

You can cast the sparse matrix to dense matrix and use tf.matmul

But if we have a large matrix with most of zero, we would like to use sparse matrix, so shouldn't cast to dense matrix

[krobasky]

I had this problem on tensorflow 2.4.1, it went away when I upgraded to 2.5 + cuda lib v11; I don't know if the trouble was with the tensorflow version, cuda, or both.