mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-06 12:20:52 +01:00
374e9373b5
Summary: Pull Request resolved: https://github.com/pytorch/pytorch/pull/44795 Today, we build our cpp tests twice, once as a standalone gtest binary, and once linked in `libtorch_python` so we can call them from `test_jit.py`. This is convenient (it means that `test_jit.py` is a single entry point for all our tests), but has a few drawbacks: 1. We can't actually use the gtest APIs, since we don't link gtest into `libtorch_python`. We're stuck with the subset that we want to write polyfills for, and an awkward registration scheme where you have to write a test then include it in `tests.h`). 2. More seriously, we register custom operators and classes in these tests. In a world where we may be linking many `libtorch_python`s, this has a tendency to cause errors with `libtorch`. So now, only tests that explicitly require cooperation with Python are built into `libtorch_python`. The rest are built into `build/bin/test_jit`. There are tests which require that we define custom classes and operators. In these cases, I've built thm into separate `.so`s that we call `torch.ops.load_library()` on. Test Plan: Imported from OSS Reviewed By: SplitInfinity, ZolotukhinM Differential Revision: D23735520 Pulled By: suo fbshipit-source-id: d146bf4e7eb908afa6f96b394e4d395d63ad72ff
86 lines
2.2 KiB
C++
86 lines
2.2 KiB
C++
#include <ATen/core/ivalue.h>
|
|
#include <c10/util/Exception.h>
|
|
#include <torch/csrc/WindowsTorchApiMacro.h>
|
|
#include <torch/csrc/jit/api/module.h>
|
|
#include <torch/script.h>
|
|
|
|
namespace torch {
|
|
namespace jit {
|
|
|
|
#ifdef _MSC_VER
|
|
#define JIT_TEST_API
|
|
#else
|
|
#define JIT_TEST_API TORCH_API
|
|
#endif
|
|
|
|
namespace {
|
|
|
|
bool isSandcastle() {
|
|
return (
|
|
(std::getenv("SANDCASTLE")) ||
|
|
(std::getenv("TW_JOB_USER") &&
|
|
std::string(std::getenv("TW_JOB_USER")) == "sandcastle"));
|
|
}
|
|
|
|
void testEvalModeForLoadedModule() {
|
|
if (isSandcastle())
|
|
return; // The module file to load is not generated in Sandcastle
|
|
std::string module_path = "dropout_model.pt";
|
|
torch::jit::Module module = torch::jit::load(module_path);
|
|
AT_ASSERT(module.attr("dropout").toModule().is_training());
|
|
module.eval();
|
|
AT_ASSERT(!module.attr("dropout").toModule().is_training());
|
|
module.train();
|
|
AT_ASSERT(module.attr("dropout").toModule().is_training());
|
|
}
|
|
|
|
void testSerializationInterop() {
|
|
if (isSandcastle()) {
|
|
// The module file to load is not generated in Sandcastle
|
|
return;
|
|
}
|
|
|
|
// This should be generated by `test/cpp/jit/tests_setup.py`
|
|
std::ifstream input_stream("ivalue.pt");
|
|
std::vector<char> input;
|
|
input.insert(
|
|
input.begin(),
|
|
std::istream_iterator<char>(input_stream),
|
|
std::istream_iterator<char>());
|
|
IValue ivalue = pickle_load(input);
|
|
|
|
auto elements = ivalue.toTuple()->elements();
|
|
auto ones = torch::ones({2, 2});
|
|
AT_ASSERT(ones.equal(elements.at(0).toTensor()));
|
|
|
|
auto twos = torch::ones({3, 5}) * 2;
|
|
AT_ASSERT(twos.equal(elements.at(1).toTensor()));
|
|
}
|
|
|
|
void testTorchSaveError() {
|
|
if (isSandcastle()) {
|
|
// The file to load is not generated in Sandcastle
|
|
return;
|
|
}
|
|
|
|
// This should be generated by `test/cpp/jit/tests_setup.py`
|
|
bool passed = true;
|
|
try {
|
|
torch::jit::load("eager_value.pt");
|
|
passed = false;
|
|
} catch (const std::exception& c) {
|
|
}
|
|
// Ensure torch::jit::load did not run
|
|
AT_ASSERT(passed);
|
|
}
|
|
} // namespace
|
|
|
|
JIT_TEST_API void runJITCPPTests() {
|
|
// TODO: this test never ran before and is broken.
|
|
// testSerializationInterop();
|
|
testEvalModeForLoadedModule();
|
|
testTorchSaveError();
|
|
}
|
|
} // namespace jit
|
|
} // namespace torch
|