|
| 1 | +# Copyright (c) Meta Platforms, Inc. and affiliates. |
| 2 | +# All rights reserved. |
| 3 | +# |
| 4 | +# This source code is licensed under the BSD-style license found in the |
| 5 | +# LICENSE file in the root directory of this source tree. |
| 6 | + |
| 7 | +# pyre-unsafe |
| 8 | + |
| 9 | +import torch |
| 10 | +import torch.nn as nn |
| 11 | +import torch.nn.functional as F |
| 12 | + |
| 13 | + |
| 14 | +class ConvlLinearModel(nn.Module): |
| 15 | + """ |
| 16 | + A neural network model with a convolutional layer followed by a linear layer. |
| 17 | + """ |
| 18 | + |
| 19 | + def __init__(self): |
| 20 | + super(ConvlLinearModel, self).__init__() |
| 21 | + self.conv_layer = nn.Conv2d( |
| 22 | + in_channels=1, out_channels=1, kernel_size=3, stride=1, padding=1 |
| 23 | + ) |
| 24 | + self.conv_layer.weight = nn.Parameter( |
| 25 | + torch.tensor([[[[0.1, 0.2, 0.3], [0.4, 0.5, 0.6], [0.7, 0.8, 0.9]]]]) |
| 26 | + ) |
| 27 | + self.conv_layer.bias = nn.Parameter(torch.tensor([0.0])) |
| 28 | + |
| 29 | + self.linear_layer = nn.Linear(in_features=4, out_features=2) |
| 30 | + self.linear_layer.weight = nn.Parameter( |
| 31 | + torch.tensor([[0.1, 0.2, 0.3, 0.4], [0.5, 0.6, 0.7, 0.8]]) |
| 32 | + ) |
| 33 | + self.linear_layer.bias = nn.Parameter(torch.tensor([0.0, 0.0])) |
| 34 | + self.additional_bias = nn.Parameter( |
| 35 | + torch.tensor([0.5, -0.5]), requires_grad=False |
| 36 | + ) |
| 37 | + self.scale_factor = nn.Parameter(torch.tensor([2.0, 0.5]), requires_grad=False) |
| 38 | + |
| 39 | + def forward(self, x): |
| 40 | + x = self.conv_layer(x) |
| 41 | + x = x.view(x.size(0), -1) |
| 42 | + x = self.linear_layer(x) |
| 43 | + x = x + self.additional_bias |
| 44 | + x = x - 0.1 |
| 45 | + x = x * self.scale_factor |
| 46 | + x = x / (self.scale_factor + 1.0) |
| 47 | + x = F.relu(x) |
| 48 | + x = torch.sigmoid(x) |
| 49 | + output1, output2 = torch.split(x, 1, dim=1) |
| 50 | + return output1, output2 |
| 51 | + |
| 52 | + @staticmethod |
| 53 | + def get_input(): |
| 54 | + """ |
| 55 | + Returns the pre-defined input tensor for this model. |
| 56 | + """ |
| 57 | + return torch.tensor([[[[1.0, 2.0], [3.0, 4.0]]]], requires_grad=True) |
| 58 | + |
| 59 | + @staticmethod |
| 60 | + def get_expected_intermediate_outputs(): |
| 61 | + """ |
| 62 | + Returns the expected outputs of the debug handles and intermediate output mapping for this model for the given input. |
| 63 | + """ |
| 64 | + return { |
| 65 | + (10,): torch.tensor([[[[7.7000, 6.7000], [4.7000, 3.7000]]]]), |
| 66 | + (11,): torch.tensor([[7.7000, 6.7000, 4.7000, 3.7000]]), |
| 67 | + (12,): torch.tensor( |
| 68 | + [ |
| 69 | + [0.1000, 0.5000], |
| 70 | + [0.2000, 0.6000], |
| 71 | + [0.3000, 0.7000], |
| 72 | + [0.4000, 0.8000], |
| 73 | + ] |
| 74 | + ), |
| 75 | + (13,): torch.tensor([[5.0000, 14.1200]]), |
| 76 | + (14,): torch.tensor([[5.5000, 13.6200]]), |
| 77 | + (15,): torch.tensor([[5.4000, 13.5200]]), |
| 78 | + (16,): torch.tensor([[10.8000, 6.7600]]), |
| 79 | + (17,): torch.tensor([3.0000, 1.5000]), |
| 80 | + (18,): torch.tensor([[3.6000, 4.5067]]), |
| 81 | + (19,): torch.tensor([[3.6000, 4.5067]]), |
| 82 | + (20,): torch.tensor([[0.9734, 0.9891]]), |
| 83 | + (21,): [torch.tensor([[0.9734]]), torch.tensor([[0.9891]])], |
| 84 | + } |
| 85 | + |
| 86 | + |
| 87 | +# Global model registry |
| 88 | +model_registry = { |
| 89 | + "ConvLinearModel": ConvlLinearModel, |
| 90 | + # Add new models here |
| 91 | +} |
| 92 | + |
| 93 | + |
| 94 | +def check_if_final_outputs_match(model_name, actual_outputs_with_handles): |
| 95 | + """ |
| 96 | + Checks if the actual outputs match the expected outputs for the specified model. |
| 97 | + Returns True if all outputs match, otherwise returns False. |
| 98 | + """ |
| 99 | + model_instance = model_registry[model_name] |
| 100 | + expected_outputs_with_handles = model_instance.get_expected_intermediate_outputs() |
| 101 | + if len(actual_outputs_with_handles) != len(expected_outputs_with_handles): |
| 102 | + return False |
| 103 | + for debug_handle, expected_output in expected_outputs_with_handles.items(): |
| 104 | + actual_output = actual_outputs_with_handles.get(debug_handle) |
| 105 | + if actual_output is None: |
| 106 | + return False |
| 107 | + if isinstance(expected_output, list): |
| 108 | + if not isinstance(actual_output, list): |
| 109 | + return False |
| 110 | + if len(actual_output) != len(expected_output): |
| 111 | + return False |
| 112 | + for actual, expected in zip(actual_output, expected_output): |
| 113 | + if not torch.allclose(actual, expected, rtol=1e-4, atol=1e-5): |
| 114 | + return False |
| 115 | + else: |
| 116 | + if not torch.allclose(actual_output, expected_output, rtol=1e-4, atol=1e-5): |
| 117 | + return False |
| 118 | + return True |
0 commit comments