vuron/adept: Платформа для обучения и вывода нейронных сетей, с API приближенным к PyTorch.

tests/grad_mode_tests.cpp

38 строк1 KB

kolkir

Add autograd disable and enable functionality

28 янв 2025, 00:19

28 янв 2025, 00:1992c91bd

#include <adept/autograd/autograd.hpp>
#include <adept/nn/cross_entropy.hpp>
#include <adept/tensor.hpp>

#include "catch.hpp"

using namespace adept;

TEST_CASE("Disable autograd backward fail", "[gradient]") {
  NoAutoGradGuard guard;
  auto x = Variable(Tensor::from_values({1.f}, {1, 1}, device_t::CPU));
  auto y = Variable(Tensor::from_values({2.f}, {1, 1}, device_t::CPU));
  auto z = x + y;
  REQUIRE_THROWS(z.backward());
}

TEST_CASE("Requires grad zero grad", "[gradient]") {
  auto x = Variable(Tensor::from_values({1.f}, {1, 1}, device_t::CPU));
  x.requires_grad(false);
  auto y = Variable(Tensor::from_values({2.f}, {1, 1}, device_t::CPU));
  auto z = x + y;

  REQUIRE_NOTHROW(z.backward());
  REQUIRE_THAT(x.grad().at<float32_t>({0, 0}), Catch::WithinRel(0.0f, 0.001f));
  REQUIRE_THAT(y.grad().at<float32_t>({0, 0}), Catch::WithinRel(1.0f, 0.001f));
}

TEST_CASE("Requires grad disable subgraph", "[gradient]") {
  auto x = Variable(Tensor::from_values({1.f, 3.f}, {1, 2}, device_t::CPU));
  auto y = Variable(Tensor::from_values({2.f, 4.f}, {1, 2}, device_t::CPU));
  auto z = x + y;
  z.requires_grad(false);
  sum(z).backward();
  REQUIRE_THAT(x.grad().at<float32_t>({0, 0}), Catch::WithinRel(0.0f, 0.001f));
  REQUIRE_THAT(y.grad().at<float32_t>({0, 0}), Catch::WithinRel(0.0f, 0.001f));
  REQUIRE_THAT(x.grad().at<float32_t>({0, 1}), Catch::WithinRel(0.0f, 0.001f));
  REQUIRE_THAT(y.grad().at<float32_t>({0, 1}), Catch::WithinRel(0.0f, 0.001f));
}