Skip to content

fix empty dropout #20541

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Closed
ngimel wants to merge 3 commits into from
Closed

fix empty dropout #20541

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
ce5c1e0
fix empty dropout
May 15, 2019
8449ea9
Fix lint
ezyang May 16, 2019
6bcbc08
Merge remote-tracking branch 'origin/fbcode/warm' into empty_dropout
ezyang May 16, 2019
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 2 additions & 2 deletions aten/src/ATen/native/Dropout.cpp
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,7 +21,7 @@ Tensor make_feature_noise(const Tensor& input) {
}

bool is_fused_kernel_acceptable(const Tensor& input, double p) {
return input.is_cuda() && p > 0 && p < 1;
return input.is_cuda() && p > 0 && p < 1 && input.numel() > 0;
}

// NB: sure, we could have used different overloads here, but I would feel insecure
Expand All @@ -41,7 +41,7 @@ Tensor multiply(const Tensor& input, const Tensor& noise) {
template<bool feature_dropout, bool alpha_dropout, bool inplace, typename T>
Ctype<inplace> _dropout_impl(T& input, double p, bool train) {
TORCH_CHECK(p >= 0 && p <= 1, "dropout probability has to be between 0 and 1, but got ", p);
if (p == 0 || !train) {
if (p == 0 || !train || input.numel() == 0) {
return input;
}

Expand Down
2 changes: 2 additions & 0 deletions aten/src/ATen/native/cuda/Dropout.cu
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -102,6 +102,8 @@ fused_dropout_cuda(const Tensor& self, double p, Generator * gen){
Tensor ret = at::empty_like(self);
Tensor mask = at::empty(self.sizes(), self.options().dtype(kByte));
const int64_t nelem = self.numel();
//empty tensors should not get here, but just in case, avoid FPE
if (nelem==0) return std::tuple<Tensor,Tensor>(self, mask);
const int64_t block_size = 256;
unsigned int blocks_per_sm = at::cuda::getCurrentDeviceProperties()->maxThreadsPerMultiProcessor/block_size;
dim3 dim_block(block_size);
Expand Down
11 changes: 11 additions & 0 deletions test/test_nn.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -5027,6 +5027,17 @@ def test_invalid_dropout_p(self):
self.assertRaises(ValueError, lambda: F.dropout(v, -0.1))
self.assertRaises(ValueError, lambda: F.dropout(v, 1.1))

def test_empty_dropout(self):
x = torch.Tensor([])
out = torch.nn.functional.dropout(x)
self.assertEqual(out.size(), x.size())

@unittest.skipIf(not TEST_CUDA, 'CUDA not available')
def test_empty_dropout_cuda(self):
x = torch.Tensor([]).to('cuda')
out = torch.nn.functional.dropout(x)
self.assertEqual(out.size(), x.size())

def test_pad_sequence(self):
def pad(tensor, length):
return torch.cat(
Expand Down