[C++ API] RNNCell / LSTMCell / GRUCell layers (#34400)

Summary:
This PR adds `RNNCell` / `LSTMCell` / `GRUCell` layers to the C++ frontend, with implementations exactly matching the Python API equivalent.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/34400

Differential Revision: D20316859

Pulled By: yf225

fbshipit-source-id: bb7cee092622334043c0d0fd0fcb4e75e707699c
This commit is contained in:
Will Feng
2020-03-13 21:49:31 -07:00
committed by Facebook GitHub Bot
parent 68758b2fa0
commit d041d0784e
7 changed files with 532 additions and 3 deletions
+30
View File
@@ -636,4 +636,34 @@ TEST_F(SequentialTest, ModuleForwardMethodOptionalArg) {
{-0.1189, 0.0502, 0.2960}}});
ASSERT_TRUE(torch::allclose(rnn_output.output, expected_output, 1e-05, 2e-04));
}
{
torch::manual_seed(0);
Sequential sequential(Identity(), RNNCell(2, 3));
auto x = torch::ones({2, 2});
auto rnn_output = sequential->forward<torch::Tensor>(x);
auto expected_output = torch::tensor(
{{-0.0645, -0.7274, 0.4531},
{-0.0645, -0.7274, 0.4531}});
ASSERT_TRUE(torch::allclose(rnn_output, expected_output, 1e-05, 2e-04));
}
{
torch::manual_seed(0);
Sequential sequential(Identity(), LSTMCell(2, 3));
auto x = torch::ones({2, 2});
auto rnn_output = sequential->forward<std::tuple<torch::Tensor, torch::Tensor>>(x);
auto expected_output = torch::tensor(
{{-0.2693, -0.1240, 0.0744},
{-0.2693, -0.1240, 0.0744}});
ASSERT_TRUE(torch::allclose(std::get<0>(rnn_output), expected_output, 1e-05, 2e-04));
}
{
torch::manual_seed(0);
Sequential sequential(Identity(), GRUCell(2, 3));
auto x = torch::ones({2, 2});
auto rnn_output = sequential->forward<torch::Tensor>(x);
auto expected_output = torch::tensor(
{{-0.1134, 0.0467, 0.2336},
{-0.1134, 0.0467, 0.2336}});
ASSERT_TRUE(torch::allclose(rnn_output, expected_output, 1e-05, 2e-04));
}
}