diff --git a/test/model/test_model.cu b/test/model/test_model.cu index 687e8ba..9169fed 100644 --- a/test/model/test_model.cu +++ b/test/model/test_model.cu @@ -8,7 +8,7 @@ TEST(Model, TestModelPredict) { int inputSize = 6; int inputChannels = 2; - int outputSize = 6; + int outputSize = 3; int kernelSize = 3; int stride = 1; @@ -26,27 +26,12 @@ TEST(Model, TestModelPredict) { ); // weights 6*6*2*2 std::vector conv2dWeights = { - 0.18313f, 0.53363f, 0.39527f, 0.27575f, 0.3433f, 0.41746f, 0.16831f, - 0.61693f, 0.54599f, 0.99692f, 0.77127f, 0.25146f, 0.4206f, 0.16291f, - 0.93484f, 0.79765f, 0.74982f, 0.78336f, 0.6386f, 0.87744f, 0.33587f, - 0.9691f, 0.68437f, 0.65098f, 0.48153f, 0.97546f, 0.8026f, 0.36689f, - 0.98152f, 0.37351f, 0.68407f, 0.2684f, 0.2855f, 0.76195f, 0.67828f, - 0.57567f, 0.6752f, 0.83416f, 0.62217f, 0.92441f, 0.96507f, 0.11171f, - 0.52438f, 0.90435f, 0.46854f, 0.59987f, 0.21747f, 0.82223f, 0.2709f, - 0.69207f, 0.16794f, 0.26679f, 0.49572f, 0.42392f, 0.49808f, 0.33058f, - 0.54071f, 0.83304f, 0.03446f, 0.65041f, 0.58601f, 0.7208f, 0.49659f, - 0.60447f, 0.70867f, 0.33336f, 0.0199f, 0.53188f, 0.15774f, 0.31791f, - 0.2611f, 0.66174f, 0.22588f, 0.95612f, 0.01209f, 0.2239f, 0.51731f, - 0.80405f, 0.09126f, 0.85215f, 0.01911f, 0.7448f, 0.61376f, 0.22161f, - 0.71483f, 0.36953f, 0.67528f, 0.77609f, 0.5933f, 0.66035f, 0.79205f, - 0.04973f, 0.78845f, 0.4023f, 0.55086f, 0.03391f, 0.06616f, 0.45673f, - 0.24497f, 0.87024f, 0.43097f, 0.24168f, 0.66931f, 0.03734f, 0.31513f, - 0.46383f, 0.74909f, 0.57525f, 0.5295f, 0.48876f, 0.89053f, 0.31964f, - 0.87242f, 0.08605f, 0.30465f, 0.8018f, 0.53419f, 0.73341f, 0.12307f, - 0.70645f, 0.40608f, 0.89397f, 0.97853f, 0.67084f, 0.47644f, 0.39974f, - 0.97879f, 0.86642f, 0.20244f, 0.66219f, 0.11623f, 0.18979f, 0.52886f, - 0.44583f, 0.41313f, 0.19766f, 0.47676f, 0.48318f, 0.02079f, 0.83777f, - 0.41167f, 0.57684f, 0.79578f, 0.17775f + 0.18313f, 0.53363f, 0.39527f, 0.27575f, 0.3433f, 0.41746f, + 0.16831f, 0.61693f, 0.54599f, 0.99692f, 0.77127f, 0.25146f, + 0.4206f, 0.16291f, 0.93484f, 0.79765f, 0.74982f, 0.78336f, + 0.6386f, 0.87744f, 0.33587f, 0.9691f, 0.68437f, 0.65098f, + 0.48153f, 0.97546f, 0.8026f, 0.36689f, 0.98152f, 0.37351f, + 0.68407f, 0.2684f, 0.2855f, 0.76195f, 0.67828f, 0.603f }; conv2d.setWeights(conv2dWeights.data()); model.addLayer("conv2d", &conv2d); @@ -59,26 +44,18 @@ TEST(Model, TestModelPredict) { // dense CUDANet::Layers::Dense dense( - 18, 6, CUDANet::Layers::ActivationType::SOFTMAX + 8, 3, CUDANet::Layers::ActivationType::SOFTMAX ); // dense weights 18*6 std::vector denseWeights = { - 0.36032f, 0.33115f, 0.02948f, 0.56265f, 0.23524f, 0.96589f, 0.09802f, - 0.45072f, 0.56266f, 0.5246f, 0.86663f, 0.30252f, 0.43514f, 0.80946f, - 0.43439f, 0.95206f, 0.5658f, 0.16344f, 0.90916f, 0.08605f, 0.07473f, - 0.95572f, 0.9127f, 0.96697f, 0.94788f, 0.66168f, 0.34927f, 0.86927f, - 0.10111f, 0.67001f, 0.09464f, 0.61963f, 0.73775f, 0.15255f, 0.37537f, - 0.72831f, 0.51559f, 0.81916f, 0.64915f, 0.23607f, 0.88699f, 0.39844f, - 0.03934f, 0.87608f, 0.68364f, 0.03633f, 0.11632f, 0.99925f, 0.86102f, - 0.6659f, 0.11022f, 0.47878f, 0.92411f, 0.38027f, 0.06771f, 0.99645f, - 0.47783f, 0.54653f, 0.41552f, 0.61055f, 0.50326f, 0.79817f, 0.20008f, - 0.32929f, 0.23562f, 0.0033f, 0.46628f, 0.04958f, 0.05235f, 0.28102f, - 0.45705f, 0.78327f, 0.91427f, 0.41122f, 0.08883f, 0.43558f, 0.14724f, - 0.74515f, 0.98215f, 0.50503f, 0.02887f, 0.25426f, 0.3463f, 0.81567f, - 0.84608f, 0.15469f, 0.6069f, 0.54311f, 0.77967f, 0.50657f, 0.18208f, - 0.7969f, 0.48401f, 0.36097f, 0.7563f, 0.50316f, 0.1134f, 0.98089f, - 0.97041f, 0.4832f, 0.79216f, 0.06572f, 0.09688f, 0.51555f, 0.1652f, - 0.73933f, 0.44365f, 0.66949f + 0.36032f, 0.33115f, 0.02948f, + 0.09802f, 0.45072f, 0.56266f, + 0.43514f, 0.80946f, 0.43439f, + 0.90916f, 0.08605f, 0.07473f, + 0.94788f, 0.66168f, 0.34927f, + 0.09464f, 0.61963f, 0.73775f, + 0.51559f, 0.81916f, 0.64915f, + 0.03934f, 0.87608f, 0.68364f, }; dense.setWeights(denseWeights.data()); model.addLayer("dense", &dense); @@ -98,12 +75,14 @@ TEST(Model, TestModelPredict) { 0.30514f, 0.23521f }; + std::vector expected = {2e-05f, 0.00021f, 0.99977f}; + // predict const float* output = model.predict(input.data()); - float sum = 0.0f; - // float sum = 0.0f; + float sum = 0.0f; for (int i = 0; i < outputSize; ++i) { + EXPECT_NEAR(expected[i], output[i], 1e-5f); sum += output[i]; } diff --git a/tools/model_test.py b/tools/model_test.py new file mode 100644 index 0000000..3373095 --- /dev/null +++ b/tools/model_test.py @@ -0,0 +1,88 @@ +import torch +import utils + +class TestModel(torch.nn.Module): + + def __init__(self, *args, **kwargs) -> None: + super().__init__(*args, **kwargs) + + self.conv1 = torch.nn.Conv2d( + in_channels=2, + out_channels=2, + kernel_size=3, + stride=1, + padding=0, + bias=False + ) + + self.maxpool1 = torch.nn.MaxPool2d( + kernel_size=2, + stride=2 + ) + self.activation = torch.nn.ReLU() + + self.linear = torch.nn.Linear( + in_features=8, + out_features=3, + bias=False + ) + self.softmax = torch.nn.Softmax(dim=0) + + def set_weights(self): + + conv2d_weights = torch.tensor([ + 0.18313, 0.53363, 0.39527, 0.27575, 0.3433, 0.41746, + 0.16831, 0.61693, 0.54599, 0.99692, 0.77127, 0.25146, + 0.4206, 0.16291, 0.93484, 0.79765, 0.74982, 0.78336, + 0.6386, 0.87744, 0.33587, 0.9691, 0.68437, 0.65098, + 0.48153, 0.97546, 0.8026, 0.36689, 0.98152, 0.37351, + 0.68407, 0.2684, 0.2855, 0.76195, 0.67828, 0.603 + + ]).reshape(2, 2, 3, 3) + self.conv1.weight = torch.nn.Parameter(conv2d_weights) + + linear_weights = torch.tensor([ + 0.36032, 0.33115, 0.02948, + 0.09802, 0.45072, 0.56266, + 0.43514, 0.80946, 0.43439, + 0.90916, 0.08605, 0.07473, + 0.94788, 0.66168, 0.34927, + 0.09464, 0.61963, 0.73775, + 0.51559, 0.81916, 0.64915, + 0.03934, 0.87608, 0.68364, + ]).reshape(3, 8) + self.linear.weight = torch.nn.Parameter(linear_weights) + + def forward(self, x): + x = self.conv1(x) + x = self.maxpool1(x) + x = self.activation(x) + + x = torch.flatten(x) + x = self.linear(x) + x = self.softmax(x) + + return x + +if __name__ == "__main__": + model = TestModel() + model.set_weights() + + input = torch.tensor([ + 0.12762, 0.99056, 0.77565, 0.29058, 0.29787, 0.58415, 0.20484, + 0.05415, 0.60593, 0.3162, 0.08198, 0.92749, 0.72392, 0.91786, + 0.65266, 0.80908, 0.53389, 0.36069, 0.18614, 0.52381, 0.08525, + 0.43054, 0.3355, 0.96587, 0.98194, 0.71336, 0.78392, 0.50648, + 0.40355, 0.31863, 0.54686, 0.1836, 0.77171, 0.01262, 0.41108, + 0.53467, 0.3553, 0.42808, 0.45798, 0.29958, 0.3923, 0.98277, + 0.02033, 0.99868, 0.90584, 0.57554, 0.15957, 0.91273, 0.38901, + 0.27097, 0.64788, 0.84272, 0.42984, 0.07466, 0.53658, 0.83388, + 0.28232, 0.48046, 0.85626, 0.04721, 0.36139, 0.6123, 0.56991, + 0.84854, 0.61415, 0.2466, 0.20017, 0.78952, 0.93797, 0.27884, + 0.30514, 0.23521 + ]).reshape(2, 6, 6) + + # input = torch.rand(2, 6, 6) + + out = model(input) + utils.print_cpp_vector(out)