experiments/STT-tensorflow
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

TF Remove the limitation of int max for dim_size in SparseSoftmax ops

Browse Source 
This fix tries to address the simmiliat issue raised in #25701 where the dim_size in SparseSoftmax ops was set to . The restriction is likely unnecessary, due to historical reasons. This fix remove this limitation.
This commit is contained in:
Dayananda-V 2019-02-13 12:20:22 +05:30
parent 126679b2e5
commit 058df02be3
1 changed files with 1 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
tensorflow/core/kernels/sparse_softmax_op.cc
View File

@ -62,12 +62,8 @@ class SparseSoftmaxOp : public OpKernel {
errors::InvalidArgument( errors::InvalidArgument(
"Input should have rank >= 2, but received shape: ", "Input should have rank >= 2, but received shape: ",
shape_t->SummarizeValue(3))); shape_t->SummarizeValue(3)));
OP_REQUIRES(context,
indices_t->dim_size(0) < std::numeric_limits<int>::max(),
errors::InvalidArgument(
"Number of non-zero elements exceeds int32 range"));
const int nnz = static_cast<int>(indices_t->dim_size(0)); const int64 nnz = indices_t->dim_size(0);
const int rank = static_cast<int>(indices_t->dim_size(1)); const int rank = static_cast<int>(indices_t->dim_size(1));
SparseTensor st; SparseTensor st;
OP_REQUIRES_OK( OP_REQUIRES_OK(