Use default dtype for torch::tensor(floating_point_values) and torch::tensor(empty braced-init-list) when dtype is not specified (#29632)

Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/29632

This PR is BC-breaking in the following way:

Previously, C++ `torch::tensor` with a floating-point literal with no suffix (e.g. `torch::tensor(1.1)`) or a (nested) braced-init-list of
floating-point literals with no suffix (e.g. `torch::tensor({{1.1, 2.2}})` produces a tensor with dtype `at::kDouble`. After this PR, it produces a tensor with dtype `torch::get_default_dtype()`, matching Python `torch.tensor` behavior.

Test Plan: Imported from OSS

Differential Revision: D18465819

Pulled By: yf225

fbshipit-source-id: 6834fe50335c677bc3832f2a5e9cf8d1ede9f665
This commit is contained in:
Will Feng
2019-11-13 15:14:08 -08:00
committed by Facebook Github Bot
parent 3fb9bbc99b
commit 2bcac59a30
9 changed files with 174 additions and 72 deletions
+3 -2
View File
@@ -29,8 +29,9 @@ void check_exact_values(
}
for (size_t p = 0; p < layerParameters.size(0); p++) {
auto tensor = layerParameters[p];
auto expectedTensor = expectedLayerParameters[p];
// Always compare using double dtype, regardless of the original dtype of the tensors
auto tensor = layerParameters[p].to(torch::kFloat64);
auto expectedTensor = expectedLayerParameters[p].to(torch::kFloat64);
if (!tensor.allclose(expectedTensor, /*rtol=*/1e-3, /*atol=*/5e-4)) {
std::cout << "layer " << i << ": " << tensor << " != " << expectedTensor